Magnetic resonance imaging characteristics of brain metastases from lung cancer

Introduction

Brain metastases (BMs) are secondary malignant tumors originating from solid organ cancers outside the central nervous system (CNS). They occur in 10% to 40% of adult patients with systemic malignant tumors and are 10 times more common than are primary malignant brain tumors. The median survival time of patients with BMs from different primary tumors is 12 months or less (1,2). Lung cancer is the most common cancer worldwide (accounting for 12.4% of all cancer cases) and the leading cause of cancer-related death (accounting for 18.7% of all cancer-related deaths), with metastatic cancer being the primary cause of cancer-related mortality (3). Lung cancer is the most common primary tumor source of BMs, accounting for approximately 50% to 60% of all BM cases (4). Non-small cell lung cancer (NSCLC) is the main type of lung cancer, constituting the majority (about 85%) of all lung cancer cases, while small-cell lung cancer (SCLC) constitutes a relatively smaller proportion (about 15%). The predominant histological subtype of NSCLC is adenocarcinoma (AD), followed by squamous cell carcinoma (SCC) (5). Although the incidence of BMs from NSCLC is lower than that from SCLC, NSCLC remains a major source of BMs, with NSCLC (especially AD) accounting for approximately 50% of BMs. SCLC is associated with a particularly high risk of BM development, with approximately 80% of patients with SCLC experiencing disease progression to BMs (2). NSCLC exhibits a high degree of molecular heterogeneity, which is reflected by its distinct genotypic subgroups. More than 60% of NSCLC cases harbor epidermal growth factor receptor (EGFR) mutations (6), which are closely associated with an increased incidence of BMs (7,8). Among patients with NSCLC, the frequency of EGFR mutations is higher in those BMs than in those without them, and patients with EGFR mutations are more prone to developing BMs than are those with wild-type EGFR (9,10). EGFR tyrosine kinase inhibitors provide high sensitivity and specificity in treating EGFR-mutant lung cancer. Consequently, they are associated with a favorable prognosis and are becoming increasingly prominent in the management of BMs (11). Therefore, timely identification of EGFR mutation status can facilitate early intervention and inform clinical decision-making, thus ultimately improving patients’ quality of life.

Magnetic resonance imaging (MRI) based on multiplanar imaging offers excellent soft-tissue resolution and superior spatial resolution. It can provide high-resolution morphological depictions of tissues’ anatomical structures and yield longitudinal information related to disease evolution and response to treatment (12,13). MRI can visualize and characterize BM lesions with unprecedented detail and is thus critical to guiding treatment strategies and providing prognostic assessment (14,15). Tumor cells capable of traversing the blood-brain barrier tend to be more invasive and metastatic, which may explain the heterogeneous distribution of MRI features for BMs, with various pathological subtypes of BMs from lung cancer exhibiting distinct phenotypic characteristics. Consequently, the MRI features of BMs can provide crucial diagnostic and therapeutic information for the individualized treatment of afflicted patients. Contrast-enhanced T1-weighted imaging (CE-T1WI) is the most routinely applied and optimal sequence for the imaging-based diagnosis and treatment of BMs. It can reflect the heterogeneous characteristics of tumor blood supply—and thus lesion heterogeneity—and possesses high clinical applicability (16). Conventional sequences, including T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and T2-fluid-attenuated inversion recovery (T2-FLAIR), provide additional morphological information on BMs and, cumulatively, are the most commonly used sequences for evaluating BMs. Conventional MRI excels in providing the detailed anatomical information of CNS tumors, while diffusion-weighted imaging (DWI), an advanced imaging technique, offers supplementary anatomical details by noninvasively detecting the movement of water molecules within living tissues (17). DWI, as the most common functional MRI method for examining the CNS, can provide high contrast between tumors and normal tissues, facilitating the detection of small lesions. Furthermore, high b-value DWI is more effective in suppressing normal tissue signals, thereby enhancing the sensitivity of tumor detection, and is therefore recommended as an effective tool for diagnosing BMs (18). In addition, given that MRI exhibits higher sensitivity in detecting BMs compared to computed tomography (CT) and positron emission tomography (19,20), MRI-based features of BMs may serve as biomarkers for the detection and monitoring of BMs.

In the management of lung cancer BMs, the clinical objective is the classification or verification of tumor subtype, as this can guide clinicians in adopting suitable therapies (21). We conducted a large-sample (1,129 patients and 8,312 lesions), multicenter (three independent centers) study on the ability of MRI features to differentiate the subtypes of lung cancer BMs. The majority of previous studies on this subject employed a single-center design with relatively small sample sizes (typically fewer than 500 cases), which limited the generalizability of their findings. By integrating the signal characteristics of DWI and CE-T1WI, we have established a clear and actionable classification framework. Unlike previous studies that relied on a single MRI sequence or non-quantitative descriptions, this seven-pattern system (patterns I–VII) potentially offers a noninvasive, reproducible tool for informing clinical decision-making. Additionally, in contrast to the limited scope of previous studies, this research systematically conducted comparative analyses across three hierarchical levels (SCLC vs. NSCLC, AD vs. SCC, and EGFR wild type vs. mutant type) to characterize the MRI patterns of BMs. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1249/rc).

Methods

Inclusion and exclusion criteria

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments, and was approved by the Ethics Committee of the Affiliated Hospital of Hebei University (approval No. HDFYLL-KY-2024-037). The requirement for written informed consent was waived due to the retrospective nature of the analysis. We conducted a retrospective analysis of clinical and MRI data from 1,129 patients diagnosed with lung cancer BM (8,312 lesions), including 369 SCLC cases (2780 lesions) and 760 NSCLC cases (5,532 lesions) from three medical centers. Among the cases, 753 (5,569 lesions) were from Center 1 (Affiliated Hospital of Hebei University; admitted between April 2016 and August 2024), 284 cases (1,666 lesions) were from Center 2 (Baoding First Central Hospital; admitted between June 2020 and June 2024), and 92 cases (1,077 lesions) were from Center 3 (The Fourth Affiliated Hospital of Hebei Medical University; admitted between September 2020 and June 2024). The inclusion criteria were as follows: (I) BM confirmed by pathology, imaging examination, or medical history; and (II) histopathological subtype of the primary lung cancer classified as SCLC, AD, or SCC based on findings from bronchoscopy, biopsy, or surgical procedures. Meanwhile, the exclusion criteria were as follows: (I) presence of other primary malignant tumors; (II) history of craniocerebral surgery; (III) incomplete clinical data of the patient; (IV) poor quality of MRI; and (V) incomplete brain MRI sequences, with any one of the five sequences (CE-T1WI, T1WI, T2WI, T2-FLAIR, or DWI) missing.

Our study was divided into three stepwise parts with distinct objectives: The aim of Part I was to use clinical MRI radiological characteristics to distinguish between BMs from NSCLC and those from SCLC. The aim of Part II was to use clinical and MRI characteristics of NSCLC BMs to distinguish between BMs from AD and those from SCC. Finally, the aim of Part III was to compare the clinical and MRI radiological characteristics of EGFR wild-type BMs with those of EGFR mutant-type BMs within NSCLC. Part I enrolled 753 cases (5,569 lesions) from Center 1, including 506 with NSCLC (3,928 lesions) and SCLC with 247 (1,641 lesions); 284 patients (1,666 lesions) from Center 2, including 183 with NSCLC (879 lesions) and 101 with SCLC (787 lesions); and 92 cases (1,077 lesions) from Center 3, including 71 with NSCLC (725 lesions) and 21 with SCLC (352 lesions). Part II (NSCLC only) enrolled 506 patients (3,928 lesions) from Center 1, including 467 with AD (3,712 lesions) and 39 with SCC (210 lesions); 183 patients (879 lesions) from Center 2, including 159 with AD (819 lesions) and 24 with SCC (60 lesions); and 71 cases (725 lesions) from Center 3, including 63 with AD (706 lesions) and 8 with SCC (19 lesions). Part III enrolled 378 cases (3,029 lesions) from Center 1, including 31 cases of the EGFR wild type (967 lesions) and 247 of the mutant type (2,062 lesions); 92 cases (453 lesions) from Center 2, including 32 of the wild type (187 lesions) and 60 of the mutant type (266 lesions); and 55 cases (655 lesions) from Center 3, including 18 of the wild type (103 lesions) and 37 of the mutant type (552 lesions). For more detailed information on the inclusion criteria and exclusion criteria, as well as information on the study procedure, please refer to Appendix 1 and Figure S1.

MRI protocol

Craniocerebral MRI was performed on either 1.5- or 3-T MRI scanners at all three centers, and DWI (b-value =1,000 s/mm²), T2-FLAIR, T2WI, and T1WI sequences, with an axial scan plane section thickness of 5 mm, were completed. Triplanar CE-T1WI sequences were obtained after weight-adjusted administration of gadodiamide-based contrast media at a dose of 0.1 mmol/kg, which was injected with a high-pressure syringe through the cubital vein at an injection rate of 2 mL/s. Imaging protocols varied across different scanners at different centers. The details of the scanning parameters of the MRI sequences are provided in Appendix 2 and Table S1.

MRI radiological features of BMs lesions

Baseline brain MRI radiological features of BM lesions were analyzed for the following variables: (I) number of BMs lesions (1, 2–5, 6–10, and >10 lesions); (II) size of BM lesions (<0.5, ≥0.5 and ≤1.0, >1.0 and ≤3.0, and >3 cm); (III) binary presence of hemorrhage or peritumoral brain edema; (IV) distribution of BMs lesions across various brain regions [supratentorial area (cortical or subcortical and deep brain) and infratentorial area]; and (V) seven patterns of DWI and CE-T1WI. MRI features were analyzed at the lesion level. For multiple lesions from the same patient, we adopted an approach of including all lesions. Since different lesions can exhibit heterogeneous signal patterns, the inclusion of all lesions allowed for a more comprehensive reflection of the association between subtypes and MRI features. Hemorrhage was defined as hyperintensity on T1WI. The sequences used for assessing peritumoral edema were T2WI and T2-FLAIR. The axial T1WI sequence was employed for determining anatomical location.

The image interpretation in this study was jointly performed by two radiologists with more than 10 years of experience in diagnostic imaging of the CNS. Prior to interpretation, all images were independently evaluated for quality by two physicians, and unqualified data with motion artifacts, insufficient contrast, etc., were excluded, and all reviewers received the same training related to the seven-pattern classification criteria. During the interpretation process, the reviewers were double-blinded to the patients’ histological subtypes and EGFR status. The interpretation workflow included a model of independent review followed by consensus resolution: the two radiologists first independently classified the lesion patterns, and for discrepant cases, a final consensus was reached through joint image review. We categorized these lesions into seven patterns based on the signal characteristics of DWI and CE-T1WI in Part I (detailed in Appendix 3 and Figures S2,S3), Part II (detailed in Appendix 3 and Figures S4,S5), and Part III (detailed in Appendix 3 and Figures S6,S7). Pattern I, characterized by DWI hypointensity/isointensity and homogeneous hyperintensity on CE-T1WI, was designated as DWI-negative plus CE-T1WI-positive. Pattern II, characterized by DWI hypointensity/isointensity and ring hyperintensity on CE-T1WI, was designated as DWI-negative plus CE-T1WI ring. Pattern III, characterized by homogeneous hyperintensity on both DWI and CE-T1WI, was designated as DWI-positive plus CE-T1WI-positive. Pattern IV, characterized by hyperintensity and homogeneous hyperintensity on CE-T1WI, was designated as DWI ring plus CE-T1WI-positive. Pattern V, characterized by DWI homogeneous hyperintensity and CE-T1WI ring hyperintensity, was designated as DWI-positive plus CE-T1WI ring. Pattern VI, characterized by ring hyperintensity on both DWI and CE-T1W, was designated as DWI ring plus CE-T1WI ring. Pattern VII, characterized by DWI homogeneous hyperintensity and CE-T1WI hypointensity/isointensity, was designated as DWI-positive plus CE-T1WI-negative. “Positive” indicates homogeneous hyperintensity on DWI or CE-T1WI, while “negative” indicates hypointensity or isointensity on DWI or CE-T1WI, and “ring” refers to ring enhancement.

Statistical analysis

The sample size was determined based on the available data, and no prior power calculation for the sample size was conducted. SPSS software version 22.0 (IBM Corp., Armonk, NY, USA) was used for data analysis. The normality of all continuous variables was verified with the Shapiro-Wilk test. Data for continuous variables conforming to a normal distribution are presented as the mean ± standard deviation; meanwhile, the data with a skewed distribution are expressed as the median with quartiles. Categorical variables are expressed as frequencies and percentages. For the analysis of the differences between the groups in the cohorts from the three medical centers and the overall cohort, the Student t-test was used for continuous variables following a normal distribution, while nonparametric tests (Mann-Whitney U test) was applied for continuous variables not following a normal distribution. Categorical variables were compared with the Pearson χ² test or the Fisher exact test. The level of significance for differences was set at P<0.05.

Results

Part I performance analysis of (SCLC vs. NSCLC)

Analysis of MRI radiological characteristics and seven patterns of BM lesions for SCLC and NSCLC

Part I included a comparative analysis of the MRI characteristics between SCLC and NSCLC BM lesions (Table 1). Regarding the number of lesions, each of the groups from centers 1–3 exhibited significances between SCLC and NSCLC BM lesions (P<0.05), but there was no difference in the overall cohort (P=0.770). In the overall cohort, both SCLC and NSCLC BMs had a relatively low proportion of single lesions, which accounted for 4.4% and 5.4% of all SCLC and NSCLC BMs, respectively, and a relatively high proportion of multiple lesions (>10 lesions) which accounted for 68.1% and 69.3% of all SCLC and NSCLC BMs, respectively. Regarding lesion size, in centers 1–3 and the overall cohort, NSCLC had a higher proportion of lesions <0.5 cm in size than did SCLC (all P values <0.001). In the overall cohort, SCLC, as compared to NSCLC, had a higher proportion of lesions ≥0.5 and ≤1.0 cm (38.8% vs. 33.7%) and >1.0 and ≤3.0 cm (29.7% vs. 19.1%) in size, and both had a relatively low proportion of lesions >3 cm in size (3.3% vs. 2.5%). Peritumoral edema differed significantly between SCLC and NSCLC in centers 2 and 3 (P<0.05), but there was no difference in Center 1 or the overall cohort (P>0.05), with SCLC and NSCLC having a similar proportion of peritumoral edema (24.4% vs. 24.1%). Hemorrhage did not differ significantly between SCLC and NSCLC in the three centers or the overall cohort (P>0.05), with a similar proportion (1.3% vs. 1.4%). The details of lesion locations are provided in Appendix 4 and Table S2.

For the seven patterns based on DWI and CE-T1WI, significant differences were found between SCLC and NSCLC across all three centers and the overall cohort (all P values <0.001). As illustrated in the grouped bar chart in Figure 1, SCLC had a higher proportion of the DWI-positive plus CE-T1WI-positive, DWI-positive plus CE-T1WI ring, DWI ring plus CE-T1WI ring, and DWI-positive plus CE-T1WI-negative patterns than did NSCLC in Center 1 (17.6% vs. 11.9%, 33.6% vs. 8.5%, 33.5% vs. 26.1%, and 10.8% vs. 4.2%, respectively), Center 2 (24.1% vs. 18.5%, 5.2% vs. 2.5%, 47.4% vs. 30.2%, and 14.7% vs. 8.1%, respectively), and the overall cohort (17.6% vs. 11.9%, 22.5% vs. 6.8%, 34.6% vs. 26.2%, and 20.6% vs. 4.5%, respectively). NSCLC had a higher proportion of the DWI-negative plus CE-T1WI-positive, DWI-negative plus CE-T1WI ring, and DWI ring plus CE-T1WI-positive patterns than did SCLC in Center 1 (37.0% vs. 2.1%, 6.7% vs. 0.4%, and 5.1% vs. 2.0%, respectively), Center 2 (26.1% vs. 2.9%, 3.0% vs. 0.9%, and 11.6% vs. 2.0%, respectively), Center 3 (40.0% vs. 0%, 27.6% vs. 2.6%, and 2.5% vs. 0%, respectively), and the overall cohort (35.9% vs. 2.1%, 8.9% vs. 0.8%, and 5.8% vs. 1.8%, respectively). The proportions of SCLC BMs lesions showing homogeneous hyperintensity on DWI in Center 1, Center 2, Center 3, and the overall cohort were 62.0%, 46.7%, 85.8%, and 60.7%, respectively. The proportions of lesions showing homogeneous enhancement, ring hyperintensity, and hypointensity/isointensity on CE-T1WI were, respectively, 17.6%, 33.6%, and 10.8% for Center 1; 24.1%, 5.2%, and 17.4% for Center 2; 3.1%, 9.1%, and 73.6% for Center 3; and 17.6%, 22.5%, and 20.6% for the overall cohort. The proportion of NSCLC BM lesions showing homogeneous hyperintensity on CE-T1WI for centers 1–3 and the overall cohort was 42.1%, 37.7%, 42.5%, and 41.7%, respectively. The proportions of lesions showing hypointensity/isointensity and ring hyperintensity on DWI were, respectively, 37.0% and 5.1% for Center 1, 26.1% and 11.6% for Center 2, 40.0% and 2.5% for Center 3, and 35.9% and 5.8% for the overall cohort. Overall, SCLC tended to present homogeneous hyperintensity on DWI and ring hyperintensity or no enhancement on CE-T1WI, whereas NSCLC tended to present homogeneous hyperintensity on CE-T1WI and ring hyperintensity on DWI.

Figure 1 Grouped bar chart showing MRI presentation patterns I–VII (DWI-negative plus CE-T1WI-positive, DWI-negative plus CE-T1WI ring, DWI-positive plus CE-T1WI-positive, DWI ring plus CE-T1WI-positive, DWI-positive plus CE-T1WI ring, DWI ring plus CE-T1WI ring, and DWI-positive plus CE-T1WI-negative) of BMs in Part I (SCLC vs. NSCLC). Center 1: Affiliated Hospital of Hebei University; Center 2: Baoding First Central Hospital; Center 3: The Fourth Affiliated Hospital of Hebei Medical University. BM, brain metastasis; CE-T1WI, contrast-enhanced T1-weighted imaging; DWI, diffusion-weighted imaging; MRI, magnetic resonance imaging; NSCLC, non-small cell lung cancer; SCLC, small-cell lung cancer.

Analysis of patients’ clinical characteristics

Part I also included a comparative analysis of the clinical characteristics of patients with BM, with the extensive findings of the univariate analysis being presented in Appendix 4 and Table S3. We observed that compared to patients with NSCLC, those with SCLC were predominantly male and smokers and were more likely to have primary lung lesions located in the left lobe, inoperable tumors, advanced tumor-node-metastasis (TNM) stage, and leptomeningeal metastases.

Part II performance analysis (AD vs. SCC)

Analysis of MRI radiological characteristics and seven patterns of BM lesions for AD and SCC

Part II included a comparative analysis of MRI characteristics of BM lesions between AD and SCC (Table 2). There was a significant difference in the number of lesions between AD and SCC for centers 2 and 3 and the overall cohort (P<0.05) but not for Center 1 (P=0.075). In the overall cohort, single lesions accounted for the smallest proportion of lesions (4.8%), while for SCC, 6–10 lesions accounted for the smallest proportion (4.2%). For both AD and SCC, >10 lesions accounted for the highest proportion (69.6% vs. 60.2%). Regarding lesion size, AD had a higher proportion of lesions 0.5–1 cm in size than did SCC for Center 2 (41.6% vs. 31.7%; P=0.004) and Center 3 (38.2% vs. 26.3%; P<0.001). In Center 2 and Center 3, both SCC and AD showed a higher proportion of lesions sized >1.0 and ≤3.0 cm, at 36.7% vs. 26.5% and 52.6% vs. 19.7%, respectively. In the overall cohort, both AD and SCC, lesions <0.5 cm in size accounted for the highest proportion (44.8% vs. 45.0%), followed by those ≥0.5 and ≤1.0 cm in size (34.1% vs. 25.3%); meanwhile, lesions >3 cm in size accounted for the smallest proportion in both AD and SCC (2.2% vs. 7.3%). The proportion of peritumoral edema differed significantly between AD and SCC in Center 1 and Center 3 and the overall cohort (P<0.05) but not in Center 2 (P=0.683). SCC had a significantly higher proportion of peritumoral edema than did AD in the overall cohort (69.9% vs. 21.6%). Hemorrhage was more common in SCC than in AD only in Center 3 (15.8% vs. 0.9%), with no significant differences in the other centers or the overall cohort; the proportion of BMs with hemorrhage was similar between SCC and ADD in the overall cohort (2.2% vs. 1.4%). The details of lesion locations are provided in Appendix 5 and Table S4.

For the seven patterns based on DWI and CE-T1WI, there was a significant difference between AD and SCC in the three centers and the overall cohort (all P values <0.001). As depicted in the grouped bar chart in Figure 2, AD had a higher proportion of the DWI-negative plus CE-T1WI-positive pattern and the DWI-negative plus CE-T1WI ring pattern than did SCC in Center 1 (39.5% vs. 1.0% and 7.0% vs. 0%, respectively), Center 2 (27.4% vs. 8.3% and 3.1% vs. 1.7%, respectively), and the overall cohort (37.8% vs. 2.4% and 9.2% vs. 2.1%, respectively). SCC had a higher proportion of the DWI ring plus CE-T1WI ring pattern than did AD in Center 1 (36.7% vs. 25.5%), Center 2 (58.3% vs. 28.1%), Center 3 (68.4% vs. 20.5%), and the overall cohort (43.3% vs. 25.3%). For AD in the overall cohort, the DWI-negative plus CE-T1WI-positive pattern was the most common, (37.8%), followed by the DWI ring plus CE-T1WI ring pattern (25.3%); meanwhile, for SCC, the DWI ring plus CE-T1WI ring patter was the most common (43.3%), followed by the DWI-positive plus CE-T1WI-negative pattern (33.9%). Overall, SCC tended to manifest as ring hyperintensity on both DWI and CE-T1WI, whereas AD tended to manifest as homogeneous hyperintensity on CE-T1WI and hypointensity or isointensity on DWI.

Figure 2 Grouped bar chart showing MRI presentation patterns I–VII (DWI-negative plus CE-T1WI-positive, DWI-negative plus CE-T1WI ring, DWI-positive plus CE-T1WI-positive, DWI ring plus CE-T1WI-positive, DWI-positive plus CE-T1WI ring, DWI ring plus CE-T1WI ring, and DWI-positive plus CE-T1WI-negative) of BMs in Part II (AD vs. SCC). Center 1: Affiliated Hospital of Hebei University; Center 2: Baoding First Central Hospital; Center 3: The Fourth Affiliated Hospital of Hebei Medical University. AD, adenocarcinoma; BM, brain metastasis; CE-T1WI, contrast-enhanced T1-weighted imaging; DWI, diffusion-weighted imaging; MRI, magnetic resonance imaging; SCC, squamous cell carcinoma.

Analysis of patients’ clinical characteristics

Part II further included a comparative analysis of the clinical characteristics between patients with AD and SCC BMs, with the details of the univariate analysis being provided in Appendix 5 and Table S5. SCC was more prevalent among older male smokers, there was a higher incidence of BMs in patients with advanced TNM stages, and the EGFR wild-type was more common.

Part III performance analysis (EGFR wild type vs. mutant type)

Analysis of MRI radiological characteristics and seven patterns of BM lesions for the EGFR wild type and mutant type

Part III included a comparative analysis of the MRI characteristics between EGFR wild-type and EGFR-mutant BM lesions (Table 3). Regarding the number of lesions, there were significant differences between the wild and mutant types in Center 2 and Center 3 and the overall cohort (P<0.05) but not in Center 1 (P=0.138). In the overall cohort, both the wild and mutant types had a relatively low proportion of single lesions (5.2% vs. 4.5%) and a relatively high proportion of multiple lesions (>10 lesions) (67.5% vs. 72.2%). Regarding lesion size, in centers >1.0 and ≤3.0 and the overall cohort, the wild type had a higher proportion of lesions >1.0 and ≤3.0 cm in size than did the mutant type (all P values <0.05). In the overall cohort, lesions <0.5 cm in size accounted for the highest proportion in both the wild and mutant types (48.9% vs. 44.0%), followed by lesions ≥0.5 and ≤1.0 cm in size (33.6% vs. 33.9%); both the wild and mutant types had a relatively low proportion of lesions >3 cm in size (2.6% vs. 1.6%). The proportion of peritumoral edema differed significantly between the wild and mutant types in Center 1 and Center 3 and the overall cohort (all P values <0.001) but not in Center 2 (P=0.692). The mutant type had a lower proportion of peritumoral edema than did the wild type in the overall cohort (15.4% vs. 26.4%). Hemorrhage did not differ significantly between the wild and mutant types in the three centers or the overall cohort, with a similar proportion between the two types (2.2% vs. 1.7%). The details regarding lesion location are provided in Appendix 6 and Table S6.

For the seven patterns based on DWI and CE-T1WI, there was a significant difference between the wild and mutant types in Center 1 (P=0.019), Center 3 (P<0.001), and the overall cohort (P<0.001) but not in Center 2 (P=0.704). As demonstrated in the grouped bar chart in Figure 3, the DWI-negative plus CE-T1WI-positive pattern accounted for the highest proportion for the both the wild and mutant types, with the mutant type having a higher proportion of this pattern compared to the wild type in Center 1 (44.0% vs. 39.7%), Center 2 (38.0% vs. 21.4%), Center 3 (44.6% vs. 40.8%), and the overall cohort (43.6% vs. 37.1%). In the overall cohort, the DWI-negative plus CE-T1WI-positive pattern accounted for the highest proportion for both the wild and mutant types (37.1% vs. 43.6%), followed by the DWI-positive plus CE-T1WI-positive (15.9% vs. 11.8%) and DWI ring plus CE-T1WI ring patterns (21.9% vs. 23.0%). Overall, BMs with the EGFR wild type group tended to exhibit homogeneous hyperintensity or ring hyperintensity on both DWI and CE-T1WI, whereas BMs with the mutant type tended to exhibit homogeneous hyperintensity on CE-T1WI and hypointensity or isointensity on DWI.

Figure 3 Grouped bar chart showing MRI presentation patterns I–VII (DWI-negative plus CE-T1WI-positive, DWI-negative plus CE-T1WI ring, DWI-positive plus CE-T1WI-positive, DWI ring plus CE-T1WI-positive, DWI-positive plus CE-T1WI ring, DWI ring plus CE-T1WI ring, and DWI-positive plus CE-T1WI-negative) of BMs in Part III (EGFR wild type vs. mutant type). Center 1: Affiliated Hospital of Hebei University; Center 2: Baoding First Central Hospital; Center 3: The Fourth Affiliated Hospital of Hebei Medical University. BM, brain metastasis; CE-T1WI, contrast-enhanced T1-weighted imaging; DWI, diffusion-weighted imaging; EGFR, epidermal growth factor receptor; MRI, magnetic resonance imaging.

Analysis of patients’ clinical characteristics

The full details of the univariate analysis of patients’ clinical characteristics from Part III are provided in Appendix 6 and Table S7. We observed that patients with mutant-type BMs tended to be male and nonsmokers and have the AD subtype.

Discussion

This study retrospectively analyzed 1,129 patients with lung cancer-related BMs from three centers, with a total of 8,312 lesions. Specifically, we examined the MRI characteristics of BMs and the corresponding clinical features of the primary lung cancer lesions. Given the significant difference in the incidence rates between SCLC and NSCLC, NSCLC is as the primary source of lung cancer-related BMs. Building upon this, the study further compared the MRI characteristics between the major pathological subtypes of NSCLC (AD vs. SCC) and EGFR wild-type status and mutant-type status. This large-scale, multicenter cohort study was conducted to overcome the limitations of single-center, small-sample studies. The seven-pattern framework based on the DWI and CE-T1WI features of BMs is an improvement upon nonuniform imaging descriptions, and the three-part, step-by-step analysis was a novel approach to achieving comprehensive subtype comparisons. The innovation of this research does not lie in the novelty of correlating MRI features of BMs with lung cancer pathology and molecular subtypes but rather in providing a reference value for the large-scale validation and practical advancement of clinically feasible MRI patterns, to a certain extent. However, it is worth noting that the associations of MRI patterns I–IV with lung cancer pathology and EGFR subtypes are “probabilistic associations” rather than “definitive imaging features”. These patterns can serve as auxiliary clinical reference indicators but should be further validated in prospective studies. Additionally, integrating them into multivariate predictive models is necessary to enhance their clinical applicability.

The findings from Part I of the analysis (SCLC vs. NSCLC) indicated that SCLC typically presents as homogeneous hyperintensity on DWI and as ring hyperintensity or non-enhancement on CE-T1WI, whereas NSCLC can be characterized by homogeneous hyperintensity on CE-T1WI and ring hyperintensity on DWI. In the subsequent Part II analysis of NSCLC (AD vs. SCC), it was found that SCC tends to exhibit ring hyperintensity on both DWI and CE-T1WI, whereas AD tends to present homogeneous hyperintensity on CE-T1WI and hypointensity/isointensity on DWI. Part III of the analysis (EGFR wild-type vs. mutant-type) suggested that wild-type BMs can be characterized by homogeneous hyperintensity or ring hyperintensity on both DWI and CE-T1WI, whereas mutant-type BMs can be characterized by homogeneous hyperintensity on CE-T1WI and hypointensity/isointensity on DWI. In summary, our findings may enhance the diagnostic sensitivity of BMs from lung cancer and aid in accurately identifying patients with neurological symptoms upon initial admission. For certain patients suspected of having lung cancer but for whom pathological diagnosis is constrained by health conditions, the combination of the clinical characteristics of lung cancer and MRI signals of BMs can serve as a diagnostic tool for differentiating between SCLC and NSCLC. Furthermore, it can facilitate the correct classification of patients with NSCLC into AD or SCC categories and aid in discerning their EGFR mutation status. However, it should be noted that the MRI patterns differed across the three centers, and the potential causes for this may be attributable to variations in equipment and scanning protocols.

In addition to assessing the ability of certain clinical characteristics to diagnosis and classify BMs from lung cancer (as detailed in Appendix 7), we applied MRI to comprehensively characterize the imaging features of each lesion in different locations in order to enhance the accuracy of differential diagnosis. Overall, a more comprehensive understanding of the MRI characteristics of lung cancer-related BMs could aid in the management of patients with uncertain or multiple primary tumors. Furthermore, a more detailed assessment of the MRI features of EGFR-mutant NSCLC BMs may better inform clinical decision-making. Our study indicates that lung cancer-related BMs tend to be multiple (with >10 lesions), have a diameter of <1 cm, involve the frontoparietal lobes in the supratentorial region and the cerebellum in the infratentorial region, and are characterized by hyperintensity on both DWI and CE-T1WI. The presence of multiple, small-volume lesions in patients with lung cancer-related BMs aligns with the fundamental characteristics of metastases.

Previous analysis suggests that the diameter of SCLC BMs is often smaller than that of NSCLC BMs (22). This finding differs from the results of our study, in which we found that in the subgroup with lesion diameters smaller than 0.5 cm, the proportion of NSCLC BMs was higher than that of SCLC BMs (44.8% vs. 28.2%; P<0.001). Larger lesions (diameters >3 cm) were characterized by intratumoral hemorrhage, which manifests as hyperintensity on T1WI. However, the proportion of BMs with diameters greater than 3 cm was small for both NSCLC and SCLC (3.3% vs. 2.5%). Consequently, intratumoral hemorrhage was not commonly observed in either group (1.3% vs. 1.4%). Additionally, peritumoral edema is often regarded as one of the significant features of BMs, typically manifesting as hyperintensity on T2WI and T2-FLAIR sequences (23). This characteristic is notably prominent in SCC BMs. Among the cases included in our study, there was a significantly higher proportion of SCC cases with peritumoral edema as compared to AD (69.9% vs. 21.6%; P<0.001). Additionally, lesions in the wild-type EGFR group exhibited a higher incidence of peritumoral edema than did those in the mutant-type group (26.4% vs. 15.4%; P<0.001). This finding aligns with the work of Huang et al., who reported that the extent of perilesional edema surrounding EGFR-mutant BMs is generally smaller compared to that observed in patients with wild-type EGFR (24). In the subgroup with lesion diameters smaller than 0.5 cm, the proportion of lesions in the mutant-type group was higher than that in the wild-type group (48.9% vs. 44.0%; P<0.001). Considering previous research findings in combination with our analysis, we can infer that EGFR-mutant BMs tend to have smaller lesion diameters and generally preserve the blood-brain barrier to a certain extent. Consequently, they are less prone to causing peritumoral edema. This suggests that the observed differences may be associated with reduced invasiveness in genetically mutated tumors. Therefore, patients with EGFR-mutant BMs may derive greater benefit from targeted therapies, which may produce more favorable prognoses (25).

According to previous studies, the location of BMs is correlated with the anatomical structure of the cranium. Nearly 80% of BMs occur in the supratentorial region, 15% in the cerebellum, and 5% in the brainstem (9). Our results indicate that lung cancer-related BMs are more commonly located in the cerebral cortex or subcortical regions and less commonly in the deeper brain areas, with the frontoparietal lobes and cerebellum being the most frequently involved sites. This may be attributed to the abundant blood supply in these anatomical regions. Furthermore, our analysis revealed that SCLC, as compared to NSCLC, is more likely to involve deeper brain regions (20.1% vs. 14.7%; P=0.002). Moreover, other less common sites, such as the brainstem and ventricles, should not be overlooked. We found that among NSCLC BMs, EGFR mutations are more prevalent in AD than in SCC (66.8% vs. 23.1%; P=0.001), which is closely associated with the higher incidence of AD. Although some research (10,26) suggests that the incidence of BMs is higher in patients with EGFR mutations than in those with wild-type EGFR, we did not find a significant difference in the distribution of BMs between patients with EGFR mutations and those without, which aligns with the observations made by Wang et al. (27). Although the distribution of lung cancer BMs may vary depending on the genetic makeup of the lung cancer, studies have also indicated that BMs with the EGFR L858R mutation are more commonly found in the caudate nucleus, cerebellum, and temporal lobe compared to those with exon 19 deletions (28). However, the correlation between the molecular mechanisms underlying the metastatic regions of BMs and EGFR status has not been definitively established. Further investigation incorporating emerging technologies such as artificial intelligence is warranted in future studies to clarify this relationship.

DWI and CE-T1WI are currently the most widely adopted and clinically practical sequences for imaging the CNS. Given that DWI reflects water diffusivity and CE-T1WI reflects the permeability of the blood-brain barrier, the analysis of signal patterns in lung cancer-related BMs based on the combination of these two modalities provides considerable clinical efficacy. DWI serves as a noninvasive marker that reflects the microscopic architecture of tissues and is influenced by tissue-specific factors such as cellular density, extracellular matrix composition, and vasogenic edema (29). Hyperintensity on DWI can aid in achieving a differential diagnosis. Additionally, the majority of lesions exhibit homogeneous enhancement. Therefore, integrating clinical evaluation with MRI analysis can provide an accurate assessment of genotype (30). CE-T1WI reflects the disruption of the blood-brain barrier and serves as a critical diagnostic indicator for BMs. However, the specific imaging characteristics of lung cancer-related BMs have not yet been definitively established. Typically, the ring hyperintensity pattern of BMs on CE-T1WI presents with sharper and more distinct margins compared with other cranial MRI sequences. Moreover, due to central necrosis, these lesions may exhibit higher cellularity and density in the surrounding regions (31). Ring hyperintensity is traditionally regarded as one of the characteristic imaging features of BMs. However, a study (32) has found that homogeneous enhancement is the most common pattern observed in SCLC BMs. Potential explanations for this observation include the relatively smaller lesion diameters, a higher number of cellular membranes within the tumor tissue, and the reduced propensity for necrosis in these lesions (32). Hyperintensity on DWI and ring hyperintensity, either as solitary features or in combination with various signal patterns, are the most commonly observed imaging characteristics in lung cancer BMs. These manifestations are likely influenced by multiple factors, including the tumor’s pathological and molecular subtypes, disease progression, and the impact of systemic therapies. Therefore, we integrated DWI and CE-T1WI to investigate the imaging disparities across different pathological subtypes of lung cancer-related BMs and the presence and absence of the EGFR mutation. Notably, multiple BMs within the same patient may exhibit distinct signal patterns, which could be attributed to variations in the timing of lesion onset and their rapid progression. We found that 60.7% of SCLC BMs had homogeneous hyperintensity on DWI and either ring hyperintensity or the absence of enhancement on CE-T1WI, a result consistent with the observations reported by Zhu et al. (32). Meanwhile, 53.6% of NSCLC BMs had homogeneous hyperintensity on CE-T1WI and ring hyperintensity on DWI. Furthermore, SCC had a higher proportion of BMs with ring hyperintensity on both DWI and CE-T1WI as compared to AD (43.3% vs. 25.3%). In contrast, AD had a higher proportion of BMs with homogeneous hyperintensity on CE-T1WI and hypointensity/isointensity on DWI as compared to SCC (37.8% vs. 2.4%). Our findings indicate that the combined imaging patterns for BMs from DWI and CE-T1WI of BMs can effectively differentiate SCLC from NSCLC and may further have the ability to subclassify NSCLC into AD and SCC. Beyond providing morphological insights into each lesion of BMs, MRI serves as a noninvasive and effective adjunctive tool to immunohistochemical analysis in the identification of metastatic tumors of uncertain origin (21). There are notable disparities in the incidence rates between SCLC and NSCLC, and oncological research has predominantly focused on the latter. This trend also holds for studies investigating BMs associated with SCLC and NSCLC. A study has indicated that the signal intensity of lung cancer-related BMs on DWI does not correlate with the tumor’s histological subtype but rather with EGFR mutation status (33). High-throughput MRI features extracted via a combined radiomics and deep learning approach have been used for the pathological classification of lung cancer-related BMs (34) and for the prediction of EGFR mutation status in NSCLC BMs (35). In our study, a higher proportion of wild-type BMs had either homogeneous hyperintensity or ring hyperintensity on both DWI and CE-T1WI as compared to mutant-type BMs (37.8% vs. 33.8%); meanwhile, a higher proportion of mutant-type BMs exhibited homogeneous hyperintensity on CE-T1WI and hypointensity/isointensity on DWI as compared to their wild-type counterparts (43.6% vs. 37.1%). Although there is currently no consensus regarding the prediction of EGFR mutation status based solely on the signal characteristics of brain MRI, our findings suggest that the early identification of critical genetic alterations may be feasible. Leveraging MRI to ascertain EGFR mutation status and accordingly tailor personalized treatment plans may provide considerable clinical utility without incurring additional costs.

In contrast to BMs that demonstrate varying degrees of enhancement and that are classified as enhancing lung cancer-related BMs, there exists a distinct subtype of non-enhancing lung cancer-related BMs identified in clinical observations. These non-enhancing BMs exhibit hyperintensity on T2WI and isointensity/hypointensity on T1WI and are devoid of surrounding edema. Additionally, they show no enhancement or only minimal ring hyperintensity on CE-T1WI (36). Based on these observations, we speculate that the absence of enhancement in such BMs may be attributed to inadequate blood supply, which subsequently impedes the infiltration of contrast agents. The permeability of blood-tumor barrier (BTB) influences the degree of enhancement in BMs. Marked enhancement suggests poor selectivity and high permeability of the BTB, whereas the lack of enhancement indicates a highly selective barrier with low permeability to traversing molecules. The inferior therapeutic response of non-enhancing BMs can be explained by their higher treatment resistance relative to enhancing BMs. Furthermore, enhancing lung cancer-related BMs and non-enhancing BMs can undergo mutual transformation during follow-up (37). Therefore, the observation of an atypical combined signal pattern characterized by hyperintensity on DWI without enhancement may represent a significant imaging feature that warrants further scrutiny. This finding further underscores the pivotal role of DWI in the detection and classification of lung cancer BMs. In our study, the non-enhancing lung cancer-related BMs, which exhibit hypointensity on T1WI and hyperintensity on T2WI, suggest a high water content within the tumor tissue, while restricted diffusion on DWI indicates a relatively dense tissue structure. This specific MRI signal pattern was more frequently observed in SCLC BMs than NSCLC BMs (20.6% vs. 4.5%; P<0.001). This finding is consistent with the results reported by Zhu et al. (32), indicating that this type of non-enhancing lung cancer-related BM is more prevalent in SCLC BMs. The hyperintensity observed on DWI in SCLC BMs is associated with the aggressive nature of SCLC, rapid tumor cell division, a higher cell population, expanded intercellular spaces, and reduced extracellular free-water mobility. Furthermore, lower apparent diffusion coefficient values are correlated with a poorer prognosis. Among NSCLC BMs, this signal pattern was more commonly encountered in SCC BMs than in AD BMs (33.9% vs. 2.8%; P<0.001), although no significant difference was noted between the EGFR wildtype and mutant-type BMs. It is also important to note that, to a certain extent, this type of BM lesion can be challenging to differentiate from lacunar acute-phase cerebral infarction, and distinguishing between them may require substantial clinical experience. Consequently, the accumulation of relevant experience is crucial for avoiding misdiagnosis in clinical practice. Additionally, we detected lesions presenting with hypointensity on DWI and hyperintensity on CE-T1WI. Although this combination of signal patterns was the most frequently observed in both the EGFR wild-type and mutant-type BMs (41.8% vs. 54.9%; P<0.001), the presence of such lesions highlights the importance of CE-T1WI, which entails a lower likelihood of misdiagnosis. Therefore, the degree of enhancement of CE-T1WI and the signal characteristics of DWI in lung cancer-related BMs can serve as effective imaging biomarkers for predicting treatment outcomes. Evaluating the enhancement degree of lung cancer-related BMs holds significant value in predicting treatment efficacy and guiding therapeutic strategies.

Despite the use of a multicenter dataset and a relatively large cohort, there were certain limitations to this study which should be acknowledged. First, due to the retrospective nature of the analysis, there was potential for selection bias and inherent challenges in adequately controlling for certain confounding variables. Lesion size may be a confounding factor in the association between MRI signal patterns (ring enhancement vs. homogeneous hyperintensity) and tumor subtypes. More specifically, smaller lesions (≤5 mm) rarely develop central necrosis, which is essential for ring-shaped hyperintensity on CE-T1WI, thus partially explaining why SCLC and SCC (with larger lesions in our cohort) showed more ring enhancement than did the AD or EGFR-mutant subtypes. Although we performed size-based subgroup analyses, residual confounding could have occurred due to partial volume effects or poor spatial resolution in tiny lesions, and future studies should adopt volumetric measurements or machine learning segmentation to address this limitation. As secondary lesions, BMs can have their MRI manifestations—including DWI/CE-T1WI patterns, lesion size, and perilesional edema—confounded by prior or concurrent systemic therapies for primary lung cancer, which constitutes another confounding variable in this study. Second, histopathological confirmation was lacking for all brain lesions, as was the survival and treatment response data that could link imaging phenotypes to outcomes. Moreover, the EGFR mutation status data were derived from primary lung cancer tissues, which may not directly reflect the secondary mutations that arise in metastatic lesions. Third, assuming lesion independence may lead to overestimation of associations in lesion-level analyses, as multiple brain metastases from the same patient often share similar characteristics. Fourth, the heterogeneity of MRI equipment and scanning parameters in our study, along with the inclusion of patients undergoing a variety of treatment and at various stages of disease, could potentially compromise the generalizability of our findings. Future multicenter studies and initiatives that integrate the CT features of pulmonary lesions with brain MRI characteristics may help enhance the practical utility of the pathological classification models for lung cancer BMs and the predictive models for EGFR gene fusions. Such endeavors, in combination with radiomics and deep learning approaches, will also facilitate further refinement of the classification-relevant MRI features of lung cancer-related BMs.

Conclusions

The combination of distinct MRI characteristics of lung cancer BMs with clinical features can serve as a noninvasive tool for clarifying the histological subtypes of patients that cannot be definitively determined through biopsy. This approach aids in differentiating between SCLC and NSCLC and may also facilitate the subclassification of NSCLC into AD and SCC. Additionally, to a certain extent, it can offer indications regarding EGFR status, thereby assisting in clinical decision-making. Interestingly, the presence of hyperintensity on DWI coupled with the absence of enhancement on CE-T1WI may cumulatively serve as a key imaging feature for classifying lung cancer BMs. This particular MRI signal pattern is more frequently observed in BMs arising from SCLC than in those arising from NSCLC. Within the spectrum of NSCLC-related brain metastases, this signal pattern is more prevalent in BMs from SCC than in those from AD, although no significant difference was noted between the EGFR wild type and mutant type.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1249/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1249/dss

Funding: This study was supported by the Postgraduate Innovation Funding Project of Hebei University (No. CXZZBS2025028) and the Hebei Key Laboratory of Precision Imaging for Inflammation-Related Tumors (No. 2363P034).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1249/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the Affiliated Hospital of Hebei University (approval No. HDFYLL-KY-2024-037). Written informed consent was waived due to its retrospective nature.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung KS. Clinical Practice Guidelines for Brain Metastasis From Solid Tumors. Brain Tumor Res Treat 2024;12:14-22. [Crossref] [PubMed]
2. Lamba N, Wen PY, Aizer AA. Epidemiology of brain metastases and leptomeningeal disease. Neuro Oncol 2021;23:1447-56. [Crossref] [PubMed]
3. Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
4. Alsabbagh R, Ahmed M, Alqudah MAY, Hamoudi R, Harati R. Insights into the Molecular Mechanisms Mediating Extravasation in Brain Metastasis of Breast Cancer, Melanoma, and Lung Cancer. Cancers (Basel) 2023;15:2258. [Crossref] [PubMed]
5. Sasaki T, Kuno H, Hiyama T, Oda S, Masuoka S, Miyasaka Y, Taki T, Nagasaki Y, Ohtani-Kim SJ, Ishii G, Kaku S, Shroff GS, Kobayashi T. 2021 WHO Classification of Lung Cancer: Molecular Biology Research and Radiologic-Pathologic Correlation. Radiographics 2024;44:e230136. [Crossref] [PubMed]
6. Soffietti R, Ahluwalia M, Lin N, Rudà R. Management of brain metastases according to molecular subtypes. Nat Rev Neurol 2020;16:557-74. [Crossref] [PubMed]
7. Shin DY, Na II, Kim CH, Park S, Baek H, Yang SH. EGFR mutation and brain metastasis in pulmonary adenocarcinomas. J Thorac Oncol 2014;9:195-9. [Crossref] [PubMed]
8. Visonà G, Spiller LM, Hahn S, Hattingen E, Vogl TJ, Schweikert G, Bankov K, Demes M, Reis H, Wild P, Zeiner PS, Acker F, Sebastian M, Wenger KJ. Machine-Learning-Aided Prediction of Brain Metastases Development in Non-Small-Cell Lung Cancers. Clin Lung Cancer 2023;24:e311-22. [Crossref] [PubMed]
9. Shukla S, Karbhari A, Rastogi S, Agarwal U, Rai P, Mahajan A. Bench-to-bedside imaging in brain metastases: a road to precision oncology. Clin Radiol 2024;79:485-500. [Crossref] [PubMed]
10. Ge M, Zhuang Y, Zhou X, Huang R, Liang X, Zhan Q. High probability and frequency of EGFR mutations in non-small cell lung cancer with brain metastases. J Neurooncol 2017;135:413-8. [Crossref] [PubMed]
11. Khandekar MJ, Piotrowska Z, Willers H, Sequist LV. Role of Epidermal Growth Factor Receptor (EGFR) Inhibitors and Radiation in the Management of Brain Metastases from EGFR Mutant Lung Cancers. Oncologist 2018;23:1054-62. [Crossref] [PubMed]
12. Derks SHAE, van der Veldt AAM, Smits M. Brain metastases: the role of clinical imaging. Br J Radiol 2022;95:20210944. [Crossref] [PubMed]
13. Dregely I, Prezzi D, Kelly-Morland C, Roccia E, Neji R, Goh V. Imaging biomarkers in oncology: Basics and application to MRI. J Magn Reson Imaging 2018;48:13-26. [Crossref] [PubMed]
14. Le Rhun E, Guckenberger M, Smits M, Dummer R, Bachelot T, Sahm F, et al. EANO-ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up of patients with brain metastasis from solid tumours. Ann Oncol 2021;32:1332-47. [Crossref] [PubMed]
15. Waizman E, Dudnik E, Lavie I, Rotem O, Amiel A, Siegal T, Haim SE, Gal O, Limon D, Tschernichovsky R, Kanner AA, Laviv Y, Katz SY. The impact of brain MRI screening on stage IV NSCLC patients: A real world look at guidelines based care. J Neurol Sci 2025;470:123398. [Crossref] [PubMed]
16. Fan B, Li M, Wang X, Xu Y, Li F, Zhang L, Jiang J, Jiang Y. Diagnostic value of gadobutrol versus gadopentetate dimeglumine in enhanced MRI of brain metastases. J Magn Reson Imaging 2017;45:1827-34. [Crossref] [PubMed]
17. Madamesila J, Tchistiakova E, Faruqi S, Das S, Ploquin N. Can machine learning models improve early detection of brain metastases using diffusion weighted imaging-based radiomics? Quant Imaging Med Surg 2023;13:7706-18. [Crossref] [PubMed]
18. Tang L, Zhou XJ. Diffusion MRI of cancer: From low to high b-values. J Magn Reson Imaging 2019;49:23-40. [Crossref] [PubMed]
19. Mărginean L, Ștefan PA, Lebovici A, Opincariu I, Csutak C, Lupean RA, Coroian PA, Suciu BA. CT in the Differentiation of Gliomas from Brain Metastases: The Radiomics Analysis of the Peritumoral Zone. Brain Sci 2022;12:109. [Crossref] [PubMed]
20. Galldiks N, Kocher M, Ceccon G, Werner JM, Brunn A, Deckert M, Pope WB, Soffietti R, Le Rhun E, Weller M, Tonn JC, Fink GR, Langen KJ. Imaging challenges of immunotherapy and targeted therapy in patients with brain metastases: response, progression, and pseudoprogression. Neuro Oncol 2020;22:17-30. [Crossref] [PubMed]
21. Bekaert L, Emery E, Levallet G, Lechapt-Zalcman E. Histopathologic diagnosis of brain metastases: current trends in management and future considerations. Brain Tumor Pathol 2017;34:8-19. [Crossref] [PubMed]
22. Wang Y, Xia W, Liu B, Zhou L, Ni M, Zhang R, Shen J, Bai Y, Weng G, Yuan S, Gao X. Exploration of spatial distribution of brain metastasis from small cell lung cancer and identification of metastatic risk level of brain regions: a multicenter, retrospective study. Cancer Imaging 2021;21:41. [Crossref] [PubMed]
23. Najafian K, Rehany B, Nowakowski A, Ghazimoghadam S, Pierre K, Zakarian R, Al-Saadi T, Reinhold C, Babajani-Feremi A, Wong JK, Guiot MC, Lacasse MC, Lam S, Siegel PM, Petrecca K, Dankner M, Forghani R. Machine learning prediction of brain metastasis invasion pattern on brain magnetic resonance imaging scans. Neurooncol Adv 2024;6:vdae200. [Crossref] [PubMed]
24. Huang Z, Tu X, Yu T, Zhan Z, Lin Q, Huang X. Peritumoural MRI radiomics signature of brain metastases can predict epidermal growth factor receptor mutation status in lung adenocarcinoma. Clin Radiol 2024;79:e305-e316. [Crossref] [PubMed]
25. Liu Y, Dai S, Liu Z, He L, Zhu L, Qin Z, Fan H, Fang F, Xie Y, Peng X. Serum tumor markers and outcomes in lung cancer patients with brain metastases: a retrospective longitudinal cohort study. Transl Lung Cancer Res 2024;13:2282-95. [Crossref] [PubMed]
26. Song SG, Kim S, Koh J, Yim J, Han B, Kim YA, Jeon YK, Chung DH. Comparative analysis of the tumor immune-microenvironment of primary and brain metastases of non-small-cell lung cancer reveals organ-specific and EGFR mutation-dependent unique immune landscape. Cancer Immunol Immunother 2021;70:2035-48. [Crossref] [PubMed]
27. Wang G, Xu J, Qi Y, Xiu J, Li R, Han M. Distribution Of Brain Metastasis From Lung Cancer. Cancer Manag Res 2019;11:9331-8. [Crossref] [PubMed]
28. Takano K, Kinoshita M, Takagaki M, Sakai M, Tateishi S, Achiha T, Hirayama R, Nishino K, Uchida J, Kumagai T, Okami J, Kawaguchi A, Hashimoto N, Nakanishi K, Imamura F, Higashiyama M, Yoshimine T. Different spatial distributions of brain metastases from lung cancer by histological subtype and mutation status of epidermal growth factor receptor. Neuro Oncol 2016;18:716-24. [Crossref] [PubMed]
29. Sanvito F, Nocera G, Wang Z, Shiuan E, Sun L, Oshima S, Kim A, Kryukov I, Shao G, Cho NS, Salamon N, Ellingson BM, Prins R, Kim W, Yao J. Diffusion and contrast-enhancement MRI phenotypes associated with immune checkpoint inhibitor exposure and with immune cell infiltration in brain metastases. Neuro Oncol 2025;27:2194-205. [Crossref] [PubMed]
30. Ren X, Zhang X, Lei X, Ma W, Zhang T, Wang Y, Ren J. Comparison of clinical and MRI features of brain metastases between ALK+ and ALK- NSCLC. Cancer Med 2024;13:e7405. [Crossref] [PubMed]
31. Dash S, Vyas S, Bhardwaj N, Ahuja CK, Modi M, Chhabra R, Sahu JK, Sankhyan N, Singh P. Synthetic MRI derived relaxometry parameters: a new insight into characterization of ring enhancing lesions of brain. Neuroradiology 2025;67:1445-54. [Crossref] [PubMed]
32. Zhu D, Shao Y, Yang Z, Cheng A, Xi Q, Liang X, Chu S. Magnetic resonance imaging characteristics of brain metastases in small cell lung cancer. Cancer Med 2023;12:15199-206. [Crossref] [PubMed]
33. Jung WS, Park CH, Hong CK, Suh SH, Ahn SJ. Diffusion-Weighted Imaging of Brain Metastasis from Lung Cancer: Correlation of MRI Parameters with the Histologic Type and Gene Mutation Status. AJNR Am J Neuroradiol 2018;39:273-9. [Crossref] [PubMed]
34. Zhang B, Zhu J, Xu R, Zou L, Lian Y, Xie X, Tian Y. A combined model integrating radiomics and deep learning based on multiparametric magnetic resonance imaging for classification of brain metastases. Acta Radiol 2025;66:24-34. [Crossref] [PubMed]
35. Cao P, Jia X, Wang X, Fan L, Chen Z, Zhao Y, Zhu J, Wen Q. Deep learning radiomics for the prediction of epidermal growth factor receptor mutation status based on MRI in brain metastasis from lung adenocarcinoma patients. BMC Cancer 2025;25:443. [Crossref] [PubMed]
36. Xu YB, Zhang Y, Song Z, Wang W, Shao L. Treatment and Prognosis of Solid and Cystic Brain Metastases in Patients with Non-Small-Cell Lung Cancer. Cancer Manag Res 2021;13:6309-17. [Crossref] [PubMed]
37. Zhang X, Yang J, Hu H, Wang Y, Zhang Z, Lv J, Li J. Longitudinal MRI evaluation of the efficacy of non-enhanced lung cancer brain metastases. Sci Rep 2025;15:3318. [Crossref] [PubMed]