A coordinate registration-based structured magnetic resonance imaging reporting method for nasopharyngeal carcinoma: a preliminary study

Introduction

Nasopharyngeal carcinoma (NPC) exhibits a distinct geographical prevalence, with over 70% of global cases occurring in China, predominantly in southern China (1). Due to the insidious nature of the lesions, about 70% of patients are diagnosed at advanced stages at the time of initial diagnosis, with a relatively low overall survival (OS), posing a significant threat to human health (2,3). A more comprehensive and accurate diagnosis of NPC is essential for improving diagnostic and therapeutic decisions and enhancing patient survival. Imaging methods, particularly magnetic resonance imaging (MRI), play a pivotal role in the clinical analysis of NPC and have received extensive research attention.

Traditional diagnostic MRI reports are written in natural language and exhibit considerable heterogeneity across centers and physicians, making it challenging to directly extract and analyze the data (4,5). Structured reports simplify medical content, reduce reading and comprehension costs by using precise terminology and scripts that cover all clinically relevant data, and can improve clinical care and workflow (6-8). Automated reporting systems for lung nodules, breast nodules, and coronary angiography have been studied to form standardized structured diagnostic reports, which greatly reduces the workload of doctors (9,10). However, due to the relative complexity and variety of anatomical structures in the head and neck region, the application of automated analysis techniques in NPC images is difficult, and the structured reporting of NPC diagnosis is time-consuming and labor-intensive for physicians (11,12). The study of structured report output through the direct interpretation and analysis of radiology images has not yet matured constrained by challenges in image segmentation, image registration, image caption generation description, etc.

In recent years, computer technology has played an increasingly important role in recognizing and processing images, which is expected to bring breakthroughs in the research of automated structured reports (13). The current NPC staging systems primarily describe tumor invasion in binary terms, lacking the details required for precise diagnosis and prognosis. In fact, detailed information about tumor invasion within anatomical structures is crucial. Studies have shown that the difference in fine anatomical structures can directly affect the treatment decision and survival prognosis of patients (14-16). Addressing this gap requires a more detailed, quantitative approach to structured reporting, which could significantly improve the accuracy and utility of NPC imaging reports.

To address these challenges, our team has developed a standardized coordinate system for nasopharyngeal MRI registration (17), providing a foundation for fine-grained analysis. This study aims to further modify this work by introducing a voxelwise invasion rate (VIR) method based on previous NPC imaging studies (18,19) to quantify tumor invasion and automate structured reporting. The objectives of this study are:

• To develop a structured reporting approach for NPC that integrates VIR analysis with a standardized coordinate system.
• To design a preliminary structured reporting framework that facilitates automated report generation.
• To provide a cost-effective, potential approach to assist clinical decision-making in NPC imaging studies.

Methods

Study design

The study was conducted in accordance with the Declaration of Helsinki (revised in 2013) and was approved by the Ethics Committee of Sun Yat-sen University Cancer Center (No. B2019-222-01). Due to the retrospective nature of this investigation, informed consent was waived. The study design is shown in Figure 1, which is divided into three main parts: tumor registration, voxel analysis, and structured report analysis.

Figure 1 Illustration of the study design. ROI, region of interest; NPC, nasopharyngeal carcinoma; VIR, voxelwise invasion rate.

First, in the tumor registration part, we have established a method based on stable anatomical landmarks for constructing a coordinate system for NPC MRI studies in our previous study (18,19). Through the six identified anatomical landmarks, we have established a stable nasopharyngeal coordinate system. Then, the tumor target regions of 778 NPC patients were registered by the nasopharyngeal coordinate system, and the tumor region of interest (ROI) were obtained. This registration method has been validated by images of the lateral pterygoid muscle (LPM), as detailed in our previously published findings (17).

Secondly, in the voxel analysis stage, we selected and outlined 20 anatomical structures ROIs relevant to the diagnosis of NPC, and these structures were subjected to voxel overlay statistical analysis with the registered tumor ROIs.

Finally, in the structured report analysis stage, we combined the patient’s VIR analysis results with the standardized nasopharyngeal coordinate system to form a structured reporting scheme. This represents an extension of our previous work, translating the image processing methodologies into a clinically applicable solution for NPC diagnosis.

Database for image structured report analysis

The MRI data in this paper were obtained from 778 patients with nonmetastatic NPC who were treated at the Cancer Center of Sun Yat-sen University from January 2010 to January 2013. The selection criteria for the data were as follows: a total of 3,814 patients with newly diagnosed NPC were enrolled; 2,973 were excluded due to incomplete medical records. Among the remaining 841 patients, 5 patients without neck MRI, 20 cases with other concurrent tumors, 24 with distant metastasis, and 14 patients without satisfactory MRI registration quality for analysis were excluded. Finally, 778 patients were included in our study. All patients completed a pretreatment examination including a complete medical history, physical examination, hematology and biochemistry profiles, chest X-ray, abdominal computed tomography, and MRI of the neck and nasopharynx (20,21). The sociodemographic and clinical characteristics of the participants are provided in Table S1.

In previous work (22,23), the Sun Yat-sen University Cancer Center has completed manual evaluation of MRIs of 778 NPC cases, including manual outlining of ROIs of primary foci, identification of 6 anatomical landmarks and manual fine reading. These works were performed by two radiologists (L.L. and L.Y., with 18 and 5 years of experience in head-and-neck cancers, respectively). For ROI outlining of primary foci and identification of the 6 anatomical landmarks in 778 NPC cases, the lesions on the axial FSE T2-weighted images were manually segmented layer-by-layer to include the entire lesion using the open-source software ITK-SNAP (www.itksnap.org/). Six anatomical landmarks were subsequently identified. These 2 tasks were done by a radiologist (L.Y.) and then validated by a senior radiologist (L.L.). For the fine reading of the images of 778 patients, the 2 radiologists reviewed the MRIs of 30 randomized cases together and then evaluated the remaining MRIs separately. The staging of each case was determined according to the 8th edition of the American Joint Committee on Cancer (AJCC) staging system (24). Disagreements were resolved by consensus.

MRIs in transverse and sagittal sections were obtained using a three-dimensional (3D) fast-recovery spin-echo sequence. MRI of patients was performed using a 3.0-T system (388 cases, Magnetom Tim Trio; Siemens Healthineers, Erlangen, Germany) or a 1.5-T system (390 cases, Signa CV/I; General Electric Healthcare, Chicago, IL, USA). The parameters of the transverse images varied between volumes, with matrices ranging from 320×320 to 768×768 pixels and slice thicknesses ranging from 4 to 6.5 mm. In each volume, 36 slices covered the entire head and neck.

To establish the coordinate system, a healthy volunteer was recruited. The parameters of the transverse section image were matrix =512×512 pixels and slice thickness =5 mm (36 slices). Meanwhile, the parameters of the sagittal section images were matrix =512×512 pixels and slice thickness =5 mm. The normal volunteer has been assessed, who has no disease or structural variation in the head and neck region, confirmed by MRI by two specialized radiologists (L.L. and L.Y.).

Tumor ROI volume registration for NPC

A coordinate system method for the nasopharynx based on anatomical landmarks has been well-validated in our earlier studies (17-19), which provided a solid foundation for the development of our registration approach. Figure 2 demonstrates a schematic of the coordinate system method based on landmarks to register the NPC patient’s tumor. Six landmarks were used in the construction of the coordinate system, including 4 landmarks right/left internal acoustic pore (RIA/LIA) and the right/left ascending segment of the internal carotid artery in the posterior cavernous sinus (RAS/LAS) in the transverse plane, and 2 landmarks the posterior clinoid process (PC) and the midpoint of the upper edge of the anterior arch of C1 (MC) in the sagittal plane. The origin of the coordinate system is located at the midpoint of the line joining the LIA and the RIA. X-axis points in the direction of the RIA. Y-axis is the direction perpendicular to the x-axis, back-to-front, in the transverse section. Z-axis is the direction perpendicular to the transverse section, top-to-bottom. The unit of each axis is in millimeters. By comparing different scanning sequences, it is known that the positional relationship markers of these anatomical points are stable and can be used as stable image registration markers.

Figure 2 Tumor registration scheme for NPC based on anatomical landmarks. RAS, right ascending segment of the internal carotid artery in the posterior cavernous sinus; LAS, left ascending segment of the internal carotid artery in the posterior cavernous sinus; RIA, right internal acoustic pore; LIA, left internal acoustic pore; PC, posterior clinoid process; MC, midpoint in the upper edge from the anterior arch of C1; NPC, nasopharyngeal carcinoma.

Once the transformation matrix of NPC patients was acquired, we were able to spatially transform the patient’s MRI and map it to a standard coordinate system to obtain a registered tumor ROI. This lays the foundation for subsequent quantitative analysis of VIR structural reports of NPC.

The selection of 20 anatomical structures

After establishing the standard coordinate system of nasopharynx, we took the template image, which was the MRI of a normal volunteer, and selected 20 typical anatomical structures of the head and neck that were of concern for the study, e.g., the levetor veli palatini muscle (LVPM), the base of the pterygoid (BP), the carotid space (CS), the cavernous sinus axial (CSA), etc.

The process of selecting the 20 anatomical structures was as follows: first, some anatomical structures related to the staging of NPC were selected by the radiologists according to the guidelines of the AJCC and their own experience. These structures represent key areas for tumor invasion and are commonly assessed in clinical practice. Then, due to the limited number of cases used in this preliminary study, structures with less than 10 invasive cases were discarded from the anatomical structures. This ROI outlining of 20 anatomical structures in normal volunteer was done by a specialized radiologist (L.Y.) and then validated by a senior radiologist (L.L.).

Table 1 lists the statistical information of the 20 typical anatomical structures we selected. In the table, True number is the number of cases in which tumor invasion was confirmed by radiologists through fine reading, indicating that the tumor has extended into the structure. Voxel number is the number of voxels that the structure occupies in 3D standardized coordinate space, which can reflect the size of the structure. These metrics form the foundation for the subsequent VIR-based structured reporting, providing quantitative data that can assist in understanding the extent of tumor invasion and structure size.

Analysis of tumor invasion of structures

Subsequently, we performed image overlap voxel detection by sequentially combining the registered NPC patient tumor ROIs with the 20 anatomical structure ROIs in the standard coordinate system. If the patient’s tumor region spatially overlapped voxels with an anatomical structure, the overlapping voxels were divided by the overall voxel of the structure to obtain the VIR of the corresponding anatomical structure for subsequent structured reports.

We defined the VIR for anatomical structures as follows:

VIR=ROItu∩ROIstrROIstr[1]

where, ROItu represents the voxels of registered tumor ROI of NPC patients, and ROIstr represents the voxels of the anatomical structure. The VIR value is easily calculated from the formula.

Figure 3 shows a schematic representation of the invasion of the LPM in different cross-sectional layers of MRI of a typical NPC patient, with the ROI of the tumor outlined in red and LPM outlined in green. In this example, the VIR value for LPM structure is obtained by calculating the ratio of the overlapping voxels between the tumor ROI (red) and the LPM ROI (green) to the total voxels of the LPM.

Figure 3 Schematic representation of tumor invasion of LPM structures in different cross-sectional layers of MRI in a patient. Outlined in red is the ROI of the tumor, and outlined in green is the ROI of the LPM structure. The z value is the z-axis coordinate value in the coordinate system. LPM, lateral pterygoid muscle; MRI, magnetic resonance imaging; ROI, region of interest.

Such a voxelwise invasion visualization can be presented to the clinicians as a structured report output to help them better determine the location, size, and risk profile of the tumor invasion. For example, if the VIR shows a high level of invasion in critical regions of the LPM, this might prompt clinicians to consider more aggressive treatment options or closer follow-up. The clinical utility of such an image output is innovative and can assist doctors in comprehensively evaluating the tumor’s behavior and making informed treatment decisions.

Analysis of VIR results for 20 anatomical structures

In total, we calculated the VIR of tumors on 20 anatomical structures for 778 cases, generating a structural report on anatomical invasion for each case. The Youden index of the receiver operating characteristic (ROC) curve was used as the classification threshold, and when the VIR was greater than the threshold, the structure was considered to be invaded, resulting in an NPC structural report output that included the invasion of each 20 anatomical structures. By comparing the VIR structured results with the clinician’s fine-reading results (25), and using VIR structured report results as the prediction value and the clinician’s fine-reading results as the true value, we can get the average accuracy of our VIR structured report for 778 cases, and based on which we propose a preliminary scheme for VIR-based fine-structured reporting.

The code for coordinate system registration and VIR analysis approach is available at https://github.com/RoxaneLiu/Lss-VIR-based-SRforNPC. Similarly, for the private dataset and other relevant parameters used in this study, interested researchers may contact us with access requests, which will be evaluated on a case-by-case basis.

Statistical analysis

A Cox regression model for OS was built using the “log-rank” parameter. P<0.05 was statistically significant.

Categorical variables were compared using Fisher’s exact test or the Chi-squared test. To estimate for interobserver agreement of the invaded structures, we randomly selected 116 cases (about 15% of the total cases) and estimated the agreement using Cohen’s kappa test. The Kappa value for diagnostic concordance was higher than 0.8.

The VIR-based intrusion results of 778 patients were compared with the results of the physician’s fine readings. The optimal cut-off value for binarization of the VIR data was determined by the Yoden index of the ROC curve. The area under the curve (AUC) was calculated to estimate the classification performance of the model. Other evaluation metrics (including sensitivity, specificity, accuracy, F1, and kappa) were also calculated separately for the above two datasets to assess the reliability of the VIR method.

All statistical analyses were performed with MATLAB (version 2021b), R software (http://www.R-project.org, version 3.4.3) and Python (https://www.python.org, version 3.11). A two-tailed P<0.05 was regarded as a significant difference.

Results

Outline of 20 anatomical structures

We used the MRI of a normal volunteer as a standard registration template, and Figure 4 shows the results of outlining 20 structures in the standard space in different cross-sections. Each color represents a structure, and due to the different sizes and locations of the structural voxels, we can observe the distribution of the 20 structures at different locations from each cross-sectional layer.

Figure 4 ROI demonstration of anatomical structures in standard space at various layers of MRI. The z value is the z-axis coordinate value in the coordinate system. Orb, orbit; ES, ethmoind sinus; CSA, cavernous sinus; SS, sphenoid sinus; Tri, trigeminal nerve; BS, base of sphenoid; PF, pterygopalatine fossa; PA, petrous apex; FOA, foramen ovale; PS, parapharyngeal space; TVPM, tenson veli palatini muscle; LC, longus capitis; LVPM, levetor veli palatini muscle; LPM, lateral pterygoid muscle; Cli, clivus; BP, base of pterygoid; MPM, medial pterygoid muscle; CS, carotid space; ROI, region of interest; MRI, magnetic resonance imaging.

Results of tumor invasion of 20 structures

We analyzed 20 structures for VIR calculations for 778 patients, and the distributional statistics of these analyses have been presented in Figure 5. Since this study is in its preliminary stage, with a limited sample of cases, and due to the scarcity of invasive samples of some of the structures, there was a high degree of serendipity in analyzing them, so only 20 typical anatomical structures were selected for study in this paper. There were significant differences in the distribution of these 20 structures. For example, structures such as the CSA, orbit (Orb) and trigeminal (Tri) have a smaller VIR. This may be due to the fact that they are predominantly distributed in the lateral part of the MRI cross-section. Only when the tumor ROI is large, it is more likely to invade these lateral structures. The distribution can show which structures are more likely to be invaded and to be invaded more extensively.

Figure 5 VIR distribution of tumor ROI on 20 anatomical structures in MRI. LVPM, levetor veli palatini muscle; CS, carotid space; LPM, lateral pterygoid muscle; LC, longus capitis; MPM, medial pterygoid muscle; PS, parapharyngeal space; TVPM, tenson veli palatini muscle; BP, base of pterygoid; BS, base of sphenoid; Cli, clivus; ES, ethmoind sinus; PA, petrous apex; SS, sphenoid sinus; CSA, cavernous sinus; Orb, orbit; Tri, trigeminal nerve; FL, foramen lacerum; FOA, foramen ovale; FRA, foramen rotundum; PF, pterygopalatine fossa; VIR, voxelwise invasion rate; ROI, region of interest; MRI, magnetic resonance imaging.

Analysis of VIR

Our image VIR structured reporting method generated a quantitative result on anatomical structure invasion for each patient. The ROC Youden index was used as the threshold (cutoff value) to binarize the VIR values, and we compared the obtained statistical results with the clinician’s fine reading results as the true standard, and obtained the accuracy rate on 20 anatomical structures. The average accuracy of the 20 structures reached 81.1%, validating the validity of the VIR structure report. More detailed evaluation metrics are summarized in Table 2.

The results indicate that the structured reporting method of VIR based on the coordinate system of the nasopharynx has been initially realized. The method can obtain the VIR of the anatomical structures in MRI, realize the structured reporting of NPC, and provide more accurate feature information for the subsequent diagnosis and treatment of NPC.

In Figure 6, an average ROC curve was generated for the 20 anatomical structures to assess the accuracy of tumor invasion detection. The AUC for 20 structures is 0.791. This value indicates that the model has moderate to good discriminatory ability in detecting voxelwise invasion. These results suggest that the proposed method can serve as a useful tool for clinicians to assess tumor invasion across various anatomical structures.

Figure 6 The average ROC curve of 20 structures. ROC, receiver operating characteristic.

A structural reporting protocol for analyzing tumor invasion

Through the VIR analysis of multiple anatomical structures of the case in the previous step, we further designed a preliminary structured reporting scheme for tumor invasion. The scheme framework is shown in Figure 7. MRIs of NPC were preprocessed to accurately outline the primary focal area, and then matched to the 3D standard space by image alignment technology to obtain the tumor ROI in the 3D standard coordinate system of the patient’s head and neck. The voxelwise overlap of the tumor ROI and the ROI of each anatomical structure after alignment were calculated to obtain the VIR, respectively. At the same time, by analyzing the data of each anatomical structure, detailed anatomical structure invasion data can be mined, including the description of the invasion position and size of each structure. The final output is a preliminary structured report based on the VIR method. To this end, we have developed an overall implementation plan and report templates to assist in the subsequent development of more comprehensive automated structured report generation. Table S2 shows a template of VIR-based structured report for NPC patients. VIR-based structured report protocol provides a new option for automated report generation, with the expectation that subsequent studies will be instructive for clinical reading outcomes and potentially help physicians standardize their reading reports.

Figure 7 Structured reporting protocol for three-dimensional tumor invasion. ROI, region of interest; VIR, voxelwise invasion rate; MRI, magnetic resonance imaging.

Discussion

In this study, we propose a structured reporting method for image VIR analysis based on the nasopharyngeal coordinate system registration, which allows for detailed assessment of tumor invasion in NPC MRIs. First, we selected 20 anatomical structures in the head and neck and calculated the VIR for each. By comparing the results with the physician’s reading, we obtained an average accuracy of 81.1% and an AUC of 0.791, demonstrating its potential for accurate quantitative analysis of tumor invasion in anatomical structures. Second, we propose a structured reporting scheme based on VIR, which is instructive for the automated generation of clinical reports. This suggests that the VIR-based structured reporting method can provide fine feature information for NPC imaging.

Overall, the main contributions of our work are as follows: (I) it establishes a VIR structured reporting method based on the nasopharyngeal coordinate system registration; (II) it demonstrates the feasibility of this method by achieving 81.1% accuracy and 0.791 AUC when compared with the manual radiologist reports; (III) it designs an initial structured reporting scheme for NPC primary foci, which can be used for subsequent automated report generation studies.

Most studies on NPC have focused on image segmentation and prognostic guidance, achieving impressive results (13,26,27). However, structured reporting methods remain underexplored, and research addressing structured clinical applications similar to our study is scarce. To the best of our knowledge, our study is the first to propose the use of VIR-based analysis for structured reporting in NPC. In contrast, our method achieved an AUC of 0.791, focusing on the integration of fine-grained voxel analysis within a structured reporting framework. This approach quantifies tumor invasion more precisely and standardizes the reporting process, offering actionable insights for clinical decision-making. This structured approach represents a significant step toward automating clinical reporting and improving diagnostic efficiency in NPC.

However, several limitations need to be addressed. First, the six anatomical landmarks and tumor ROIs used for registration were manually outlined by clinicians, and the study included data from only 778 NPC patients, which may limit the generalizability of the results. The manual outlining process is resource-intensive, and it is necessary to investigate automated outlining schemes. In recent years, studies have integrated machine learning, particularly deep learning techniques, with MRI to achieve accurate and stable automatic ROI segmentation of NPC primary foci (28-30). Expanding the dataset and incorporating automated segmentation techniques could improve the study’s statistical validity and applicability to a broader range of clinical settings.

In addition, this study is in the preliminary stage, follow-up should be need to further improve the formation of a comprehensive structural report method. Limited by the problem of the number of samples, this study only covered 20 structures in the transverse plane, future work will need to include structures from the coronal and sagittal planes, such as Atlas which is usually analyzed in the sagittal plane. Furthermore, our initial reporting protocol primarily addresses primary foci but is yet to consider lymph node involvement, which is critical for accurate N-staging analysis and prognosis. In our future work, we will combine imaging histology and clinical multimodal information (31,32) to refine our structured reporting system and explore its role in clinical decision-making and prognosis.

The main clinical significance of this study lies in integrating VIR-based quantitative analysis into structured reporting for clinical practice, addressing the limitations of current manual narrative reports. By automating tumor invasion detection and providing standardized, quantifiable data, this approach can improve the efficiency of diagnosis and treatment planning. The method could be incorporated into existing clinical workflows, for example, a program can be developed by integrating automated tumor ROI identification, VIR calculation, and structured report output using the method described in this article. Once MRI examination is finished, images can be directly transferred in to the program, and then an automated structured report can be acquired, which would be verified by a radiologist. This would reduce radiologists’ workload on report generation, and thus allowing them to focus more on evaluation and interpretation of the case rather than manually writing reports. Additionally, by offering standardized and reproducible reports, this method can help improve consistency between clinicians, which is critical for tumor staging and treatment decision-making.

Conclusions

In this study, we performed head and neck structured analysis of MRIs of 778 NPC patients based on the nasopharyngeal coordinate system with anatomical landmarks and constructed a VIR structured reporting of tumors. Our proposed VIR structured reporting method achieved an average accuracy of 81.1% and an average AUC of 0.791 on 20 structures. This proved the effectiveness of the method in the analysis of NPC invasion. Through VIR analysis, we have successfully designed a preliminary structured reporting scheme for tumor invasion, which provides a new automated method for structured reporting of NPC. Our results preliminarily realized the structured reporting of VIR based on the nasopharyngeal coordinate system, which provides a new perspective and methodology for the imaging study of NPC, and is expected to provide physicians with more accurate and refined diagnostic and therapeutic decision-making support in future clinical practice, with rich clinical application prospects.

Acknowledgments

None.

Footnote

Funding: The study was partially supported by project grants from the National Natural Science Foundation of China (Nos. 82260358 and 82171906), and the Scientific Research Basic Ability Enhancement Program for Young and Middle-aged Teachers of Guangxi (No. 2024KY0202).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-24-1127/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. The study was conducted in accordance with the Declaration of Helsinki (revised in 2013) and approved by the Ethics Committee of Sun Yat-sen University Cancer Center (No. B2019-222-01). Due to the retrospective nature of this investigation, informed consent was waived.

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. Wang Y, Yan Q, Fan C, Mo Y, Wang Y, Li X, Liao Q, Guo C, Li G, Zeng Z, Xiong W, Huang H. Overview and countermeasures of cancer burden in China. Sci China Life Sci 2023;66:2515-26. [Crossref] [PubMed]
2. Goshtasbi K, Lehrich BM, Birkenbeuel JL, Abiri A, Harris JP, Kuan EC. A Comprehensive Analysis of Treatment Management and Survival Outcomes in Nasopharyngeal Carcinoma. Otolaryngol Head Neck Surg 2021;165:93-103. [Crossref] [PubMed]
3. Liu H, Tang L, Li Y, Xie W, Zhang L, Tang H, Xiao T, Yang H, Gu W, Wang H, Chen P. Nasopharyngeal carcinoma: current views on the tumor microenvironment's impact on drug resistance and clinical outcomes. Mol Cancer 2024;23:20. [Crossref] [PubMed]
4. Gore A, Hoch MJ, Shu HG, Olson JJ, Voloschin AD, Weinberg BD. Institutional Implementation of a Structured Reporting System: Our Experience with the Brain Tumor Reporting and Data System. Acad Radiol 2019;26:974-80. [Crossref] [PubMed]
5. Bhayana R, Nanda B, Dehkharghanian T, Deng Y, Bhambra N, Elias G, Datta D, Kambadakone A, Shwaartz CG, Moulton CA, Henault D, Gallinger S, Krishna S. Large Language Models for Automated Synoptic Reports and Resectability Categorization in Pancreatic Cancer. Radiology 2024;311:e233117. [Crossref] [PubMed]
6. Stanzione A, Boccadifuoco F, Cuocolo R, Romeo V, Mainenti PP, Brunetti A, Maurea S. State of the art in abdominal MRI structured reporting: a review. Abdom Radiol (NY) 2021;46:1218-28. [Crossref] [PubMed]
7. Hong SB, Lee SS, Kim JH, Kim HJ, Byun JH, Hong SM, Song KB, Kim SC. Pancreatic Cancer CT: Prediction of Resectability according to NCCN Criteria. Radiology 2018;289:710-8. [Crossref] [PubMed]
8. Adams LC, Truhn D, Busch F, Kader A, Niehues SM, Makowski MR, Bressem KK. Leveraging GPT-4 for Post Hoc Transformation of Free-text Radiology Reports into Structured Reporting: A Multilingual Feasibility Study. Radiology 2023;307:e230725. [Crossref] [PubMed]
9. Abidi SA, Hoch MJ, Hu R, Sadigh G, Voloschin A, Olson JJ, Shu HG, Neill SG, Weinberg BD. Using Brain Tumor MRI Structured Reporting to Quantify the Impact of Imaging on Brain Tumor Boards. Tomography 2023;9:859-70. [Crossref] [PubMed]
10. Nobel JM, van Geel K, Robben SGF. Structured reporting in radiology: a systematic review to explore its potential. Eur Radiol 2022;32:2837-54. [Crossref] [PubMed]
11. Wong LM, Ai QYH, Poon DMC, Tong M, Ma BBY, Hui EP, Shi L, King AD. A convolutional neural network combined with positional and textural attention for the fully automatic delineation of primary nasopharyngeal carcinoma on non-contrast-enhanced MRI. Quant Imaging Med Surg 2021;11:3932-44. [Crossref] [PubMed]
12. Ye Y, Cai Z, Huang B, He Y, Zeng P, Zou G, Deng W, Chen H, Huang B. Fully-Automated Segmentation of Nasopharyngeal Carcinoma on Dual-Sequence MRI Using Convolutional Neural Networks. Front Oncol 2020;10:166. [Crossref] [PubMed]
13. Ng WT, But B, Choi HCW, de Bree R, Lee AWM, Lee VHF, López F, Mäkitie AA, Rodrigo JP, Saba NF, Tsang RKY, Ferlito A. Application of Artificial Intelligence for Nasopharyngeal Carcinoma Management - A Systematic Review. Cancer Manag Res 2022;14:339-66. [Crossref] [PubMed]
14. Li S, Zhang W, Liang B, Huang W, Luo C, Zhu Y, Kou KI, Ruan G, Liu L, Zhang G, Li H. A Rulefit-based prognostic analysis using structured MRI report to select potential beneficiaries from induction chemotherapy in advanced nasopharyngeal carcinoma: A dual-centre study. Radiother Oncol 2023;189:109943. [Crossref] [PubMed]
15. Huang W, Quan T, Zhao Q, Li S, Cai Y, Zhou J, Luo C, Ruan G, Cui C, Liang S, Li H, Liu L. MRI of nasopharyngeal carcinoma: parapharyngeal subspace involvement has prognostic value and influences T-staging in the IMRT era. Eur Radiol 2022;32:262-71. [Crossref] [PubMed]
16. Zhu S, Li S, Cao D, Luo C, Liang Z, Liang S, Zhang G, Zhao Q, Ruan G, Liu L, Fu G, Li H. Prognostic Value of Skull Base Foramen Invasion Subclassification in T Category Modification and Induction Chemotherapy Management for Nasopharyngeal Carcinoma: Post-Hoc Analysis of a Dual-Center Retrospective Cohort Study. Adv Sci (Weinh) 2025;12:e2408182. [Crossref] [PubMed]
17. Chen H, Li H, Yang S, Huang W, Gong Q, Ruan G, Chen S, Liu L. Prognostic potential of a voxelwise invasion risk map of nasopharyngeal carcinoma based on a coordinate system of the nasopharynx. Quant Imaging Med Surg 2023;13:982-98. [Crossref] [PubMed]
18. Li S, Gong Q, Li H, Chen S, Liu Y, Ruan G, Zhu L, Liu L, Chen H. Automatic location scheme of anatomical landmarks in 3D head MRI based on the scale attention hourglass network. Comput Methods Programs Biomed 2022;214:106564. [Crossref] [PubMed]
19. Li S, Chen S, Li H, Ruan G, Ren S, Zhang T, Liu L, Chen H. Anatomical Point-of-Interest Detection in Head MRI Using Multipoint Feature Descriptor. IEEE Access 2020;8:173239-49.
20. Cui C, Wang S, Zhou J, Dong A, Xie F, Li H, Liu L. Machine Learning Analysis of Image Data Based on Detailed MR Image Reports for Nasopharyngeal Carcinoma Prognosis. Biomed Res Int 2020;2020:8068913. [Crossref] [PubMed]
21. Huang L, Zhang Y, Liu Y, Li H, Wang S, Liang S, Zhou J, Cui C, Sun Y, Chen M, Xu S, Li J, Liu L. Prognostic value of retropharyngeal lymph node metastasis laterality in nasopharyngeal carcinoma and a proposed modification to the UICC/AJCC N staging system. Radiother Oncol 2019;140:90-7. [Crossref] [PubMed]
22. Liu Y, Chen S, Dong A, Ai F, Quan T, Cui C, Zhou J, Liang S, Wang J, Wang S, Hua L, Xu S, Chen M, Sun Y, Li H, Liu L. Nodal grouping in nasopharyngeal carcinoma: prognostic significance, N classification, and a marker for the identification of candidates for induction chemotherapy. Eur Radiol 2020;30:2115-24. [Crossref] [PubMed]
23. Li HJ, Hu YY, Huang L, Zhou J, Li JJ, Xie CB, Chen MY, Sun Y, Liu LZ, Tian L. Subclassification of skull-base invasion for nasopharyngeal carcinoma using cluster, network and survival analyses: A double-center retrospective investigation. Radiother Oncol 2019;134:37-43. [Crossref] [PubMed]
24. Pan JJ, Mai HQ, Ng WT, Hu CS, Li JG, Chen XZ, et al. Ninth Version of the AJCC and UICC Nasopharyngeal Cancer TNM Staging Classification. JAMA Oncol 2024;10:1627-35. [Crossref] [PubMed]
25. Dong A, Huang W, Ma H, Cui C, Zhou J, Ruan G, Liang S, Liu L, Li H. Grading Soft Tissue Involvement in Nasopharyngeal Carcinoma Using Network and Survival Analyses: A Two-Center Retrospective Study. J Magn Reson Imaging 2021;53:1752-63. [Crossref] [PubMed]
26. OuYang PY, He Y, Guo JG, Liu JN, Wang ZL, Li A, Li J, Yang SS, Zhang X, Fan W, Wu YS, Liu ZQ, Zhang BY, Zhao YN, Gao MY, Zhang WJ, Xie CM, Xie FY. Artificial intelligence aided precise detection of local recurrence on MRI for nasopharyngeal carcinoma: a multicenter cohort study. EClinicalMedicine 2023;63:102202. [Crossref] [PubMed]
27. Luo X, Liao W, He Y, Tang F, Wu M, Shen Y, Huang H, Song T, Li K, Zhang S, Zhang S, Wang G. Deep learning-based accurate delineation of primary gross tumor volume of nasopharyngeal carcinoma on heterogeneous magnetic resonance imaging: A large-scale and multi-center study. Radiother Oncol 2023;180:109480. [Crossref] [PubMed]
28. Men K, Chen X, Yang B, Zhu J, Yi J, Wang S, Li Y, Dai J. Automatic segmentation of three clinical target volumes in radiotherapy using lifelong learning. Radiother Oncol 2021;157:1-7. [Crossref] [PubMed]
29. Tang P, Zu C, Hong M, Yan R, Peng X, Xiao J, Wu X, Zhou J, Zhou L, Wang Y. DA-DSUnet: Dual Attention-Based Dense SU-Net for Automatic Head-and-Neck Tumor Segmentation in MRI Images. Neurocomputing 2021;435:103-13.
30. Yang X, Wu J, Chen X. Application of Artificial Intelligence to the Diagnosis and Therapy of Nasopharyngeal Carcinoma. J Clin Med 2023;12:3077. [Crossref] [PubMed]
31. Yan H, Xiao S, Fu S, Gong J, Qi Z, Chen G, Chen P, Tang G, Su T, Yang Z, Wang Y. Functional and structural brain abnormalities in substance use disorder: A multimodal meta-analysis of neuroimaging studies. Acta Psychiatr Scand 2023;147:345-59. [Crossref] [PubMed]
32. Rajpurkar P, Lungren MP. The Current and Future State of AI Interpretation of Medical Images. N Engl J Med 2023;388:1981-90. [Crossref] [PubMed]