Insights into Imaging最新文献

{"title":"Radiology AI and sustainability paradox: environmental, economic, and social dimensions.","authors":"Burak Kocak, Andrea Ponsiglione, Valeria Romeo, Lorenzo Ugga, Merel Huisman, Renato Cuocolo","doi":"10.1186/s13244-025-01962-2","DOIUrl":"https://doi.org/10.1186/s13244-025-01962-2","url":null,"abstract":"<p><p>Artificial intelligence (AI) is transforming radiology by improving diagnostic accuracy, streamlining workflows, and enhancing operational efficiency. However, these advancements come with significant sustainability challenges across environmental, economic, and social dimensions. AI systems, particularly deep learning models, require substantial computational resources, leading to high energy consumption, increased carbon emissions, and hardware waste. Data storage and cloud computing further exacerbate the environmental impact. Economically, the high costs of implementing AI tools often outweigh the demonstrated clinical benefits, raising concerns about their long-term viability and equity in healthcare systems. Socially, AI risks perpetuating healthcare disparities through biases in algorithms and unequal access to technology. On the other hand, AI has the potential to improve sustainability in healthcare by reducing low-value imaging, optimizing resource allocation, and improving energy efficiency in radiology departments. This review addresses the sustainability paradox of AI from a radiological perspective, exploring its environmental footprint, economic feasibility, and social implications. Strategies to mitigate these challenges are also discussed, alongside a call for action and directions for future research. CRITICAL RELEVANCE STATEMENT: By adopting an informed and holistic approach, the radiology community can ensure that AI's benefits are realized responsibly, balancing innovation with sustainability. This effort is essential to align technological advancements with environmental preservation, economic sustainability, and social equity. KEY POINTS: AI has an ambivalent potential, capable of both exacerbating global sustainability issues and offering increased productivity and accessibility. Addressing AI sustainability requires a broad perspective accounting for environmental impact, economic feasibility, and social implications. By embracing the duality of AI, the radiology community can adopt informed strategies at individual, institutional, and collective levels to maximize its benefits while minimizing negative impacts.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"88"},"PeriodicalIF":4.1,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12006592/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144020711","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A deep learning model based on self-supervised learning for identifying subtypes of proliferative hepatocellular carcinoma from dynamic contrast-enhanced MRI.","authors":"Hui Qu, Shuairan Zhang, Xuedan Li, Yuan Miao, Yuxi Han, Ronghui Ju, Xiaoyu Cui, Yiling Li","doi":"10.1186/s13244-025-01968-w","DOIUrl":"https://doi.org/10.1186/s13244-025-01968-w","url":null,"abstract":"<p><strong>Objectives: </strong>This study employs dynamic contrast-enhanced MRI (DCE-MRI) to noninvasively predict the proliferative subtype of hepatocellular carcinoma (HCC). This subtype is marked by high tumor proliferation and aggressive clinical behavior. We developed a deep learning prediction model that employs a dynamic radiomics workflow and self-supervised learning (SSL). The model analyzes temporal and spatial patterns in DCE-MRI data to identify the proliferative subtype efficiently and accurately. Our goal is to improve diagnostic precision and guide personalized treatment planning.</p><p><strong>Methods: </strong>This retrospective study included 381 HCC patiephonnts who underwent curative resection at two medical centers. The cohort was divided into the training (n = 220), internal (n = 93), and external (n = 68) test sets. A DL model was developed using DCE-MRI of the primary tumor. Class activation mapping was used to interpret HCC proliferation in HCC.</p><p><strong>Results: </strong>The pHCC-SSL model performed well in predicting HCC proliferation, with a training set AUC) of 1.00, an internal test set AUC of 0.91, and an external test set AUC of 0.94. Without SSL pre-training, the AUC for internal and external testing decreased to 0.81 and 0.80, respectively. The predictive performance of the derived model was superior to that of the current single-sequence model.</p><p><strong>Conclusions: </strong>The pHCC-SSL model employs dynamic radiomics and a two-stage training approach to efficiently predict HCC proliferation from multi-sequence DCE-MRI, surpassing traditional single-stage models in accuracy and speed.</p><p><strong>Critical relevance statement: </strong>Our study introduces the pHCC-SSL model, a self-supervised deep learning approach using DCE-MRI that enhances the diagnostic accuracy of HCC subtypes, significantly advancing clinical radiology by enabling personalized treatment strategies.</p><p><strong>Key points: </strong>The proposed model enables noninvasive identification of HCC with high proliferation and aggressive behavior. SSL improves lesion differentiation by reducing redundancy and enhancing feature diversity. Dynamic feature extraction captures vascular infiltration, aiding preoperative metastasis risk assessment.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"89"},"PeriodicalIF":4.1,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12006648/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143984892","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Asclepius with contemporary glasses introduces 'Modern Radiology'-an open-access teaching resource.","authors":"Minerva Becker","doi":"10.1186/s13244-025-01953-3","DOIUrl":"https://doi.org/10.1186/s13244-025-01953-3","url":null,"abstract":"<p><p>The European Society of Radiology (ESR) initially developed the ESR e-Book as an open-access digital teaching resource aligned with the undergraduate European Training Curriculum for Radiology. Over time, this initiative evolved into Modern Radiology, a rebranded and expanded version that now serves not only medical students and educators but also residents and other healthcare professionals. Unlike traditional textbooks, Modern Radiology follows a visually engaging and succinct format, emphasising key concepts through annotated images, attention points, hyperlinks, and self-assessment sections. It complements case-based and interactive learning methods, providing a systematic foundation for radiology education. By continuously updating its content to reflect advancements in the field, Modern Radiology has been widely adopted by universities across Europe and beyond. Furthermore, several translations are in progress, ensuring its accessibility to a broader audience. KEY POINTS: Modern Radiology is an open-source, non-commercial teaching resource for undergraduate and postgraduate radiology teaching. It is a living document that follows the European Society of Radiology (ESR) Training Curriculum for Radiology. Its didactic and concise format facilitates the acquisition of basic knowledge and the preparation of interactive teaching and self-learning. It is widely used in universities across Europe and beyond, as well as by residents in training.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"87"},"PeriodicalIF":4.1,"publicationDate":"2025-04-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12003229/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143963318","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Classification and segmentation of hip fractures in x-rays: highlighting fracture regions for interpretable diagnosis.","authors":"Germán González, Joaquín Galant, José María Salinas, Emilia Benítez, Maria Dolores Sánchez-Valverde, Jorge Calbo, Nicolás Cerrolaza","doi":"10.1186/s13244-025-01958-y","DOIUrl":"https://doi.org/10.1186/s13244-025-01958-y","url":null,"abstract":"<p><strong>Objective: </strong>To develop an artificial intelligence (AI) system capable of classifying and segmenting femoral fractures. To compare its performance against existing state-of-the-art methods.</p><p><strong>Methods: </strong>This Institutional Review Board (IRB)-approved retrospective study did not require informed consent. 10,308 hip x-rays from 2618 patients were retrieved from the hospital PACS. 986 were randomly selected for annotation and randomly split into training, validation, and test sets at the patient level. Two radiologists segmented and classified femoral fractures based on their location (femoral neck, pertrochanteric region, or subtrochanteric region) and grade, using the Evans and Garden scales for neck and pertrochanteric regions, respectively. A YOLOv8 segmentation convolutional neural network (CNN) was trained to generate fracture masks and indicate their class and grade. Classification CNNs were trained in the same dataset for method comparison.</p><p><strong>Results: </strong>On the test set, YOLOv8 achieved a Dice coefficient of 0.77 (95% CI: 0.56-0.98) for segmenting fractures, an accuracy of 86.2% (95% CI: 80.77-90.55) for classification and grading, and an AUC of 0.981 (95% CI: 0.965-0.997) for fracture detection. These metrics are on par with or exceed those of previously published AI methods, demonstrating the efficacy of our approach.</p><p><strong>Conclusions: </strong>The high accuracy and AUC values demonstrate the potential of the proposed neural network as a reliable tool in clinical settings. Further, it is the first to provide a precise segmentation of femoral fractures, as indicated by the Dice scores, which may enhance interpretability. A formal evaluation is planned to further assess its clinical applicability.</p><p><strong>Critical relevance statement: </strong>The proposed system offers high granularity in fracture classification and is the first to segment femoral fractures, ensuring interpretability.</p><p><strong>Key points: </strong>We present the first AI method that segments and grades femoral fractures. The method classifies fractures with fracture location and type. High accuracy and interpretability promise utility in clinical practice.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"86"},"PeriodicalIF":4.1,"publicationDate":"2025-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12000489/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144021768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Correction: Fluoroscopy-guided aspiration of the acutely dislocated total hip arthroplasty: a feasible, high-yield, and safe procedure.","authors":"Dyan V Flores, Abdullah Felemban, Taryn Hodgdon, Paul Beaulé, George Grammatopoulos, Kawan S Rakhra","doi":"10.1186/s13244-025-01963-1","DOIUrl":"https://doi.org/10.1186/s13244-025-01963-1","url":null,"abstract":"","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"85"},"PeriodicalIF":4.1,"publicationDate":"2025-04-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11996738/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143985417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI-based automatic estimation of single-kidney glomerular filtration rate and split renal function using non-contrast CT.","authors":"Yiwei Wang, Feng Xu, Qiuyue Han, Daoying Geng, Xin Gao, Bin Xu, Wei Xia","doi":"10.1186/s13244-025-01959-x","DOIUrl":"10.1186/s13244-025-01959-x","url":null,"abstract":"<p><strong>Objectives: </strong>To address SPECT's radioactivity, complexity, and costliness in measuring renal function, this study employs artificial intelligence (AI) with non-contrast CT to estimate single-kidney glomerular filtration rate (GFR) and split renal function (SRF).</p><p><strong>Methods: </strong>245 patients with atrophic kidney or hydronephrosis were included from two centers (Training set: 128 patients from Center I; Test set: 117 patients from Center II). The renal parenchyma and hydronephrosis regions in non-contrast CT were automatically segmented by deep learning. Radiomic features were extracted and combined with clinical characteristics using multivariable linear regression (MLR) to obtain a radiomics-clinical-estimated GFR (rcGFR). The relative contribution of single-kidney rcGFR to overall rcGFR, the percent renal parenchymal volume, and the percent renal hydronephrosis volume were combined by MLR to generate the estimation of SRF (rcphSRF). The Pearson correlation coefficient (r), mean absolute error (MAE), and Lin's concordance coefficient (CCC) were calculated to evaluate the correlations, differences, and agreements between estimations and SPECT-based measurements, respectively.</p><p><strong>Results: </strong>Compared to manual segmentation, deep learning-based automatic segmentation could reduce the average segmentation time by 434.6 times to 3.4 s. Compared to single-kidney GFR measured by SPECT, the rcGFR had a significant correlation of r = 0.75 (p < 0.001), MAE of 10.66 mL/min/1.73 m², and CCC of 0.70. Compared to SRF measured by SPECT, the rcphSRF had a significant correlation of r = 0.92 (p < 0.001), MAE of 7.87%, and CCC of 0.88.</p><p><strong>Conclusions: </strong>The non-contrast CT and AI methods are feasible to estimate single-kidney GFR and SRF in patients with atrophic kidney or hydronephrosis.</p><p><strong>Critical relevance statement: </strong>For patients with an atrophic kidney or hydronephrosis, non-contrast CT and artificial intelligence methods can be used to estimate single-kidney glomerular filtration rate and split renal function, which may minimize the radiation risk, enhance diagnostic efficiency, and reduce costs.</p><p><strong>Key points: </strong>Renal function can be assessed using non-contrast CT and AI. Estimated renal function significantly correlated with the SPECT-based measurements. The efficiency of renal function estimation can be refined by the proposed method.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"84"},"PeriodicalIF":4.1,"publicationDate":"2025-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11977085/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143795076","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Reply to the Letter to the Editor: Should all trainees \"do research\"?","authors":"Luis Martí-Bonmatí","doi":"10.1186/s13244-025-01954-2","DOIUrl":"10.1186/s13244-025-01954-2","url":null,"abstract":"","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"80"},"PeriodicalIF":4.1,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11971094/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143788296","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Renal angiomyolipoma-investigating radiological signs indicative of risk for bleeding.","authors":"Jesper Swärd, Karl Bohlin, Olof Henrikson, Sven Lundstam, Ralph Peeker, Anna Grenabo Bergdahl","doi":"10.1186/s13244-025-01957-z","DOIUrl":"10.1186/s13244-025-01957-z","url":null,"abstract":"<p><strong>Objectives: </strong>To compare imaging differences between bleeding and non-bleeding angiomyolipoma with respect to the proportion and attenuation of the angiomyogenic component and the occurrence and size of aneurysms.</p><p><strong>Materials and methods: </strong>CT scans and angiographies preceding 58 consecutive embolisations at two institutions from 1999 to 2018 were analysed retrospectively. Tumour volume was measured by contouring the angiomyolipoma on CT scans. The partial volume of the angiomyogenic component (blood vessels and smooth muscle relative to fatty tissue) was derived using attenuation threshold values measured in Hounsfield Units.</p><p><strong>Results: </strong>Bleeding angiomyolipoma exhibited a significantly higher proportion of angiomyogenic component (23%) than non-bleeding angiomyolipoma (8%) (p = 0.042). Angiomyolipoma with 0-5% angiomyogenic component had a lower risk of bleeding compared to those with ≥ 5% angiomyogenic component (13% vs 42%). Mean attenuation values of angiomyogenic components did not differ between bleeders and non-bleeders. Aneurysms were observed in 24% of angiomyolipoma during angiography. No statistically significant association was found between the occurrence of aneurysms and bleeding, neither when all aneurysms were included nor when only aneurysms ≥ 5 mm were considered. Tuberous sclerosis patients had larger tumours (11.4 cm vs 6.0 cm), but no significant difference in bleeding was observed (p = 0.53).</p><p><strong>Conclusions: </strong>A higher proportion of the angiomyogenic component in bleeding renal angiomyolipoma suggests a possible association with bleeding. Angiomyolipoma with less than 5% angiomyogenic components may represent a subgroup with a reduced risk of bleeding. Our findings do not confirm the widely accepted assumption that aneurysms significantly increase the risk of bleeding.</p><p><strong>Critical relevance statement: </strong>Measuring the angiomyogenic component in renal angiomyolipoma could help address current knowledge gaps and aid in the more efficient selection of patients for therapeutic interventions.</p><p><strong>Key points: </strong>Identifying risk factors for bleeding beyond tumour size is important. Very low angiomyogenic component tumours may have reduced bleeding risk. The presence of aneurysms may not significantly increase bleeding risk. Reporting angiomyogenic proportion on CT may aid in treatment decisions.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"83"},"PeriodicalIF":4.1,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11972250/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143788273","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Quality of prostate MRI in early diagnosis-a national survey and reading evaluation.","authors":"Linda C P Thijssen, Jasper J Twilt, Tristan Barrett, Francesco Giganti, Ivo G Schoots, Rianne R M Engels, Mireille J M Broeders, Jelle O Barentsz, Maarten de Rooij","doi":"10.1186/s13244-025-01960-4","DOIUrl":"10.1186/s13244-025-01960-4","url":null,"abstract":"<p><strong>Objectives: </strong>The reliability of image-based recommendations in the prostate cancer pathway is partially dependent on prostate MRI image quality. We evaluated the current compliance with PI-RADSv2.1 technical recommendations and the prostate MRI image quality in the Netherlands. To aid image quality improvement, we identified factors that possibly influence image quality.</p><p><strong>Materials and methods: </strong>A survey was sent to 68 Dutch medical centres to acquire information on prostate MRI acquisition. The responding medical centres were requested to provide anonymised prostate MRI examinations of biopsy-naive men suspected of prostate cancer. The images were evaluated for quality by three expert prostate radiologists. The compliance with PI-RADSv2.1 technical recommendations and the PI-QUALv2 score was calculated. Relationships between hardware, education of personnel, technical parameters, and/or patient preparation and both compliance and image quality were analysed using Pearson correlation, Mann-Whitney U-test, or Student's t-test where appropriate.</p><p><strong>Results: </strong>Forty-four medical centres submitted their compliance with PI-RADSv2.1 technical recommendations, and 26 medical centres completed the full survey. Thirteen hospitals provided 252 usable images. The mean compliance with technical recommendations was 79%. Inadequate PI-QUALv2 scores were given in 30.9% and 50.6% of the mp-MRI and bp-MRI examinations, respectively. Multiple factors with a possible relationship with image quality were identified.</p><p><strong>Conclusion: </strong>In the Netherlands, the average compliance with PI-RADSv2.1 technical recommendations is high. Prostate MRI image quality was inadequate in 30-50% of the provided examinations. Many factors not covered in the PI-RADSv2.1 technical recommendations can influence image quality. Improvement of prostate MRI image quality is needed.</p><p><strong>Critical relevance statement: </strong>It is essential to improve the image quality of prostate MRIs, which can be achieved by addressing factors not covered in the PI-RADSv2.1 technical recommendations.</p><p><strong>Key points: </strong>Prostate MRI image quality influences the diagnostic accuracy of image-based decisions. Thirty to fifty percent of Dutch prostate MRI examinations were of inadequate image quality. We identified multiple factors with possible influence on image quality.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"82"},"PeriodicalIF":4.1,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11972232/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143788173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CT-based radiomics deep learning signatures for non-invasive prediction of metastatic potential in pheochromocytoma and paraganglioma: a multicohort study.","authors":"Yongjie Zhou, Yuan Zhan, Jinhong Zhao, Linhua Zhong, Fei Zou, Xuechao Zhu, Qiao Zeng, Jiayu Nan, Lianggeng Gong, Yongming Tan, Lan Liu","doi":"10.1186/s13244-025-01952-4","DOIUrl":"10.1186/s13244-025-01952-4","url":null,"abstract":"<p><strong>Objectives: </strong>This study aimed to develop and validate CT-based radiomics deep learning signatures for the non-invasive prediction of metastatic potential in pheochromocytomas and paragangliomas (PPGLs).</p><p><strong>Methods: </strong>We conducted a retrospective analysis of 249 PPGL patients from three institutions, dividing them into training (n = 138), test1 (n = 71), and test2 (n = 40) sets. Based on the grading system for adrenal pheochromocytoma and paraganglioma (GAPP), patients were classified into low-risk (GAPP < 3) and high-risk (GAPP ≥ 3) groups. Radiomic features were extracted from CT venous phase images and modeled using six machine learning algorithms. The maximum 2D sections and 3D images of each tumor were input into four ResNet models to obtain predictive probabilities. Optimal models were selected based on receiver operating characteristic analysis and integrated with radiological features to develop a combined model, which was evaluated on external datasets, and explored prognostic information.</p><p><strong>Results: </strong>The support vector machine radiomics and 2D ResNet-50 models demonstrated good performance. By integrating these two models with intratumoral necrosis features, we constructed a combined model that achieved high accuracy, with area under the curve (AUC) values of 0.90 for the training, 0.86 for the test1, and 0.88 for the test2 sets. This model effectively stratified patients based on metastasis-free survival (p = 0.003). Its predictive ability remains robust below the 6 cm threshold, with AUC values exceeding 0.87 across all datasets.</p><p><strong>Conclusions: </strong>The combined model can predict the metastatic potential of PPGL in the preoperative stage, providing a precise surgical strategy for pheochromocytoma regarding the 6 cm surgical threshold.</p><p><strong>Critical relevance statement: </strong>The combined model, established based on radiomic and deep learning signatures, shows potential for early preoperative prediction of metastatic potential in PPGL.</p><p><strong>Key points: </strong>Metastatic potential of PPGL affects surgical approaches and prognosis. CT-based radiomics deep learning signatures can predict the metastatic potential in PPGL.3. The combined model's predictive ability remains robust below the 6-cm threshold. The combined model's predictive ability remains robust below the 6-cm threshold.</p>","PeriodicalId":13639,"journal":{"name":"Insights into Imaging","volume":"16 1","pages":"81"},"PeriodicalIF":4.1,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11971077/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143788112","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}