Abdominal Radiology最新文献

{"title":"Deep learning assisted detection and segmentation of uterine fibroids using multi-orientation magnetic resonance imaging","authors":"Xin-Yu Liu,&nbsp;Zhi-Lin Yuan,&nbsp;Fu-Ze Cong,&nbsp;Li Mao,&nbsp;Xiu-Li Li,&nbsp;Zhen Zhou,&nbsp;Jing Ren,&nbsp;Yuan Li,&nbsp;Yan Zhang,&nbsp;Yong-Lan He,&nbsp;Hua-Dan Xue,&nbsp;Zheng-Yu Jin","doi":"10.1007/s00261-025-04934-8","DOIUrl":"10.1007/s00261-025-04934-8","url":null,"abstract":"<div><h3>Purpose</h3><p>To develop deep learning models for automated detection and segmentation of uterine fibroids using multi-orientation MRI.</p><h3>Methods</h3><p>Pre-treatment sagittal and axial T2-weighted MRI scans acquired from patients diagnosed with uterine fibroids were collected. The proposed segmentation models were constructed based on the three-dimensional nnU-Net framework. Fibroid detection efficacy was assessed, with subgroup analyses by size and location. The segmentation performance was evaluated using Dice similarity coefficients (DSCs), 95% Hausdorff distance (HD95), and average surface distance (ASD).</p><h3>Results</h3><p>The internal dataset comprised 299 patients who were divided into the training set (n = 239) and the internal test set (n = 60). The external dataset comprised 45 patients. The sagittal T2WI model and the axial T2WI model demonstrated recalls of 74.4%/76.4% and precision of 98.9%/97.9% for fibroid detection in the internal test set. The models achieved recalls of 93.7%/95.3% for fibroids ≥ 4 cm. The recalls for International Federation of Gynecology and Obstetrics (FIGO) type 2–5, FIGO types 012(submucous), fibroids FIGO types 567(subserous) were 100%/100%, 73.3%/78.6%, and 80.3%/81.9%, respectively. The proposed models demonstrated good performance in segmentation of the uterine fibroids with mean DSCs of 0.789 and 0.804, HD95s of 9.996 and 10.855 mm, and ASDs of 2.035 and 2.115 mm in the internal test set, and with mean DSCs of 0.834 and 0.818, HD95s of 9.971 and 11.874 mm, and ASDs of 2.031 and 2.273 mm in the external test set.</p><h3>Conclusion</h3><p>The proposed deep learning models showed promise as reliable methods for automating the detection and segmentation of the uterine fibroids, particularly those of clinical relevance.</p><h3>Graphical abstract</h3>\u0000<div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4927 - 4938"},"PeriodicalIF":2.2,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143787633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Endometriosis MDC: role of the radiologist","authors":"Bryan Buckley,&nbsp;Zeyad Elias,&nbsp;Garvit Khatri,&nbsp;Scott Young,&nbsp;Leann Kania,&nbsp;Priyanka Jha,&nbsp;Anuradha Shenoy-Bhangle,&nbsp;Ania Kielar","doi":"10.1007/s00261-025-04920-0","DOIUrl":"10.1007/s00261-025-04920-0","url":null,"abstract":"<div><p>Endometriosis presents a common and significant health burden affecting approximately 1 in 10 reproductive age patients who are assigned female at birth. Recently guidelines have begun shifting away from laparoscopy as the first-line diagnostic tool and instead recommend imaging for initial diagnosis and presurgical mapping of endometriosis. This shift places the radiologist in a central role for diagnosis and assessment of management efforts. During this comprehensive review article, we discuss the importance of MDC in the treatment of advanced endometriosis, the radiologist ‘s role at endometriosis MDC as well as the opportunities that exist for the radiologist to add-value in an endometriosis service.</p></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4914 - 4926"},"PeriodicalIF":2.2,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143778734","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Arterial phase CT radiomics for non-invasive prediction of Ki-67 proliferation index in pancreatic solid pseudopapillary neoplasms","authors":"Jun Liu,&nbsp;Huanhua Wu,&nbsp;Dabin Ren,&nbsp;Hao Huang,&nbsp;Xinyue Chen,&nbsp;Liqiu Liu,&nbsp;Yongtao Wang,&nbsp;Guoyu Wang","doi":"10.1007/s00261-025-04921-z","DOIUrl":"10.1007/s00261-025-04921-z","url":null,"abstract":"<div><h3>Background</h3><p>This study aimed to preoperatively predict Ki-67 proliferation levels in patients with pancreatic solid pseudopapillary neoplasm (pSPN) using radiomics features extracted from arterial phase helical CT images.</p><h3>Methods</h3><p>We retrospectively analyzed 92 patients (Ningbo Medical Center Lihuili Hospital: <i>n</i> = 64, Taizhou Central Hospital: <i>n</i> = 28) with pathologically confirmed pSPN from June 2015 to June 2023. Ki-67 positivity &gt; 3% was considered high. Radiomics features were extracted using PyRadiomics, with patients from training cohort (<i>n</i> = 64) and validation cohort (<i>n</i> = 28). A radiomics signature was constructed, and a CT radiomics score (CTscore) was calculated. Deep learning models were employed for prediction, with early stopping to prevent overfitting.</p><h3>Results</h3><p>Seven key radiomics features were selected via LASSO regression with cross-validation. The deep learning model demonstrated improved accuracy with demographics and CTscore, with key features such as Morphology and CTscore contributing significantly to predictive accuracy. The best-performing models, including GBM and deep learning algorithms, achieved high predictive performance with an AUC of up to 0.946 in the training cohort.</p><h3>Conclusions</h3><p>We developed a robust deep learning-based radiomics model using arterial phase CT images to predict Ki-67 levels in pSPN patients, identifying CTscore and Morphology as key predictors. This non-invasive approach has potential utility in guiding personalized preoperative treatment strategies.</p><h3>Clinical trial number</h3><p>Not applicable.</p></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4635 - 4645"},"PeriodicalIF":2.2,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143771008","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Clinico-radiological attributes of abnormal pancreaticobiliary junction","authors":"Shravya Bhargavi Dontheneni,&nbsp;Aasritha Kotha,&nbsp;Tharani Putta,&nbsp;Shashank Chapala,&nbsp;Suvarna Naidu Nagipagu","doi":"10.1007/s00261-025-04917-9","DOIUrl":"10.1007/s00261-025-04917-9","url":null,"abstract":"<div><h3>Aims</h3><p>To evaluate the prevalence of abnormal pancreaticobiliary junction (APBJ) on Magnetic Resonance Cholangio-Pancreatography (MRCP) in patients with and without choledochal cyst (CDC), and study their clinico-radiological profile.</p><h3>Methods</h3><p>We have retrospectively screened all MRCP studies (<i>n</i> = 13,482) done in our Radiology department over 18 months and documented the presence and type of APBJ (any length of extra-duodenal common channel) and CDC, other co-existing pancreaticobiliary abnormalities including complications.</p><h3>Results</h3><p>Prevalence of APBJ was 0.5% (<i>n</i> = 67) with 77% of them showing CDC (52/67) while only 0.85% of patients without APBJ have CDC (p value &lt; 0.0001). The most common type of CDC associated with APBJ was Todani Type I (86%) followed by type IV (14%). 31% of CDC patients had APBJ (52 out of 165) while the majority of patients with CDC (69%) did not have APBJ. Between the CDC (<i>n</i> = 52) and non-CDC (<i>n</i> = 15) subgroups of APBJ, there was statistically significant difference in the age (25 vs. 40 years, p value 0.003), gender, length of common channel (14.4 <i>±</i> 6 mm vs. 10.6 <i>±</i> 5 mm, p value 0.03), JSPBM type of APBJ and the risk of biliary malignancy (1.9% vs. 26.7%, p value 0.008, Odds ratio 13.8). Although idiopathic pancreatitis was also more common in the non-CDC subgroup, this difference was not statistically significant. There was no statistical correlation between the length of common channel and occurrence of CDC, biliary calculi, malignancy or pancreatitis.</p><h3>Conclusion</h3><p>Any length of common pancreaticobiliary channel outside the duodenal wall must be considered as APBJ; there is no correlation between the actual length of common channel and occurrence of its complications. The often overlooked and underdiagnosed subgroup of APBJ without biliary dilatation are 13.8 times more likely to develop biliary malignancy than the CDC group. We therefore suggest a necessary shift in surveillance strategies and advocate for routine screening of patients with APBJ for any biliary malignancy, even in the absence of CDC, and perhaps subject them to prophylactic cholecystectomy.</p><h3>Graphical Abstract</h3><div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4625 - 4634"},"PeriodicalIF":2.2,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143771009","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predicting clinically significant prostate cancer in PI-RADS 3 lesions using MRI-based radiomics: a literature review of methodological variations and performance","authors":"Alejandro Serrano,&nbsp;Christopher Louviere,&nbsp;Anmol Singh,&nbsp;Savas Ozdemir,&nbsp;Mauricio Hernandez,&nbsp;K. C. Balaji,&nbsp;Dheeraj R. Gopireddy,&nbsp;Kazim Z. Gumus","doi":"10.1007/s00261-025-04914-y","DOIUrl":"10.1007/s00261-025-04914-y","url":null,"abstract":"<div><h3>Purpose</h3><p>To evaluate the current state of MRI-based radiomics for predicting clinically significant prostate cancer (csPCa) in PI-RADS 3 lesions and assess the quality of these radiomic studies via a systematic review of the published literature.</p><h3>Methods</h3><p>We conducted a literature search in PubMed, EMBASE, and SCOPUS databases from January 2017 to September 2024, using search terms containing variations of PI-RADS-3 and radiomics in abstract and titles. We collected details from the radiomic workflow for each study, including statistical performance of the radiomics models (area under the curve (AUC)). We calculated the pooled AUC across the studies and a radiomics quality score (RQS) to evaluate the quality of radiomics methodology.</p><h3>Results</h3><p>Of 52 articles retrieved, 14 met the selection criteria. Of these, 12 studies employed 3T MRI scanners, 8 studies T2WI, DWI, ADC images for feature extraction, and 13 studies performed manual segmentation. All but two studies used the <i>PyRadiomics</i> platform as their feature extraction tool. The most commonly used radiomic selection methods were Least Absolute Shrinkage and Selection Operator (LASSO). The total number of features extracted ranged between 107 and 2553. The median number of radiomics features selected for use in models was 10. Nine studies (9/14) explored clinical variables in their radiomics models, with the most common being age and PSA. For building the final model, Logistic Regression, and Univariate and Multivariate modeling methods were featured across eight studies (8/14). Overall performance of the models by pooled AUC was 0.823 (95% CI, 0.72, 0.92). The mean RQS score was 15/36 (range 13–19).</p><h3>Conclusion</h3><p>MRI-based radiomic models have potential in predicting csPCa in PI-RADS-3 lesions. However, using RQS as a guide, we determined there is a clear need to improve the methodological quality of existing and future studies by focusing on extensive validation and open publishing of data for reproducibility.</p><h3>Graphical Abstract</h3><div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4783 - 4795"},"PeriodicalIF":2.2,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143762742","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Splenic imaging and non-vascular intervention: when to intervene on the spleen and how to do it safely","authors":"Danielle E. Kruse,&nbsp;Benjamin Wildman-Tobriner,&nbsp;Lisa M. Ho","doi":"10.1007/s00261-025-04925-9","DOIUrl":"10.1007/s00261-025-04925-9","url":null,"abstract":"<div><p>Cross-sectional interventional radiology (CSIR) for the spleen has evolved over the past two decades. Contemporary data show that CT- and US-guided splenic procedures are safer than once assumed, allowing for biopsy and drainage when clinically appropriate. Prior to intervention, however, close imaging evaluation of splenic lesions is required in order to recognize definitively benign, “do not touch” lesions as well as suspicious features that might warrant biopsy. This manuscript will first focus on imaging workup and decision-making for splenic lesions, and then discuss modern CSIR techniques to maximize both safety and yield.</p></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"5019 - 5027"},"PeriodicalIF":2.2,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143762728","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transverse testicular ectopia with persistent müllerian duct syndrome: a case report","authors":"Keisuke Hidaka,&nbsp;Akihiro Furuta,&nbsp;Nobuyuki Mori,&nbsp;Takuya Maekura,&nbsp;Fuki Shitano,&nbsp;Yang Wang,&nbsp;Naoko Nishio,&nbsp;Yumiko Fujiwara,&nbsp;Toshiya Takamura,&nbsp;Kotoe Hidaka,&nbsp;Toshiki Shiozaki","doi":"10.1007/s00261-025-04915-x","DOIUrl":"10.1007/s00261-025-04915-x","url":null,"abstract":"<div><p>Transverse testicular ectopia (TTE) is a variant of ectopic testis in which both testes are located in the same scrotum. Persistent Müllerian duct syndrome (PMDS) is a congenital anomaly characterized by the retention of Müllerian duct remnants in males. TTE is a characteristic clinical finding in PMDS. This case report describes a patient in his 50s who was diagnosed with PMDS based on TTE findings and a pelvic structure suggestive of a Müllerian duct remnant. Computed tomography (CT) and magnetic resonance imaging (MRI) revealed the normal left testis within the left scrotum. The soft tissue was identified dorsal to the left testis, supplied from branches of bilateral internal iliac arteries. Additionally, a rod-like structure was observed in the left pelvis, appearing continuous with the soft tissue in the left scrotum. The right testis, enlarged and torsed, was located cranially within the left scrotum. A cord-like structure extended continuously from this ectopically positioned right testis toward the right inguinal region. Surgical findings revealed a normal left testis and torsion of the right testis. The right testis was removed. Histological examination revealed a seminoma within the right testis, and immunostaining confirmed the presence of attached Müllerian duct remnants. Prior imaging revealed that the right testis was located in the abdominal cavity, indicating that TTE in this case did not occur during testicular descent. In PMDS cases, TTE is believed to result from the traction of one testis by Müllerian duct remnants and the testicular mobility due to an abnormally long gubernaculum, which is a crucial role in normal testicular descent. The findings in this case support this theory and further indicate that TTE can occur even after testicular descent is complete. This report describes a case of TTE with a causative Müllerian duct remnant and a presumed abnormally long gubernaculum identified on imaging, which contributes to the understanding of this rare condition.</p><h3>Graphical abstract</h3><div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4796 - 4802"},"PeriodicalIF":2.2,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143762730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep learning-based segmentation of gallbladder cancer on abdominal computed tomography scans: a multicenter study","authors":"Pankaj Gupta,&nbsp;Niharika Dutta,&nbsp;Ajay Tomar,&nbsp;Shravya Singh,&nbsp;Sonam Choudhary,&nbsp;Nandita Mehta,&nbsp;Vansha Mehta,&nbsp;Rishabh Sheth,&nbsp;Divyashree Srivastava,&nbsp;Salai Thanihai,&nbsp;Palki Singla,&nbsp;Gaurav Prakash,&nbsp;Thakur Yadav,&nbsp;Lileswar Kaman,&nbsp;Santosh Irrinki,&nbsp;Harjeet Singh,&nbsp;Niket Shah,&nbsp;Amit Choudhari,&nbsp;Shraddha Patkar,&nbsp;Mahesh Goel,&nbsp;Rajnikant Yadav,&nbsp;Archana Gupta,&nbsp;Ishan Kumar,&nbsp;Kajal Seth,&nbsp;Usha Dutta,&nbsp;Chetan Arora","doi":"10.1007/s00261-025-04887-y","DOIUrl":"10.1007/s00261-025-04887-y","url":null,"abstract":"<div><h3>Objectives</h3><p>To train and validate segmentation models for automated segmentation of gallbladder cancer (GBC) lesions from contrast-enhanced CT images.</p><h3>Materials and methods</h3><p>This retrospective study comprised consecutive patients with pathologically proven treatment naïve GBC who underwent a contrast-enhanced CT scan at four different tertiary care referral hospitals. The training and validation cohort comprised CT scans of 317 patients (center 1). The internal test cohort comprised a temporally independent cohort (<i>n</i> = 29) from center 1 (internal test 1). The external test cohort comprised CT scans from three centers [ (<i>n</i> = 85)]. We trained the state-of-the-art 2D and 3D image segmentation models, SAM Adapter, MedSAM, 3D TransUNet, SAM-Med3D, and 3D-nnU-Net, for automated segmentation of the GBC. The models’ performance for GBC segmentation on the test datasets was assessed via dice score and intersection over union (IoU) using manual segmentation as the reference standard.</p><h3>Results</h3><p>The 2D models performed better than 3D models. Overall, MedSAM achieved the highest dice and IoU scores on both the internal [mean dice (SD) 0.776 (0.106) and mean IoU 0.653 (0.133)] and external [mean dice (SD) 0.763 (0.098) and mean IoU 0.637 (0.116)] test sets. Among the 3D models, TransUNet showed the best segmentation performance with mean dice (SD) and IoU (SD) of 0.479 (0.268) and 0.356 (0.235) in the internal test and 0.409 (0.339) and 0.317 (0.283) in the external test sets. The segmentation performance was not associated with GBC morphology. There was weak correlation between the dice/IoU and the size of the GBC lesions for any segmentation model.</p><h3>Conclusion</h3><p>We trained 2D and 3D GBC segmentation models on a large dataset and validated these models on external datasets. MedSAM, a 2D prompt-based foundational model, achieved the best segmentation performance.</p><h3>Graphical abstract</h3><div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4615 - 4624"},"PeriodicalIF":2.2,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143750226","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The “laminated or swirled” sign","authors":"Sitthipong Srisajjakul,&nbsp;Patcharin Prapaisilp,&nbsp;Sirikan Bangchokdee","doi":"10.1007/s00261-025-04913-z","DOIUrl":"10.1007/s00261-025-04913-z","url":null,"abstract":"","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"5064 - 5065"},"PeriodicalIF":2.2,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143750434","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Attention mechanism-based multi-parametric MRI ensemble model for predicting tumor budding grade in rectal cancer patients","authors":"Jianye Jia,&nbsp;Yue Kang,&nbsp;Jiahao Wang,&nbsp;Fan Bai,&nbsp;Lei Han,&nbsp;Yantao Niu","doi":"10.1007/s00261-025-04886-z","DOIUrl":"10.1007/s00261-025-04886-z","url":null,"abstract":"<div><h3>Purpose</h3><p>To develop and validate a deep learning-based feature ensemble model using multiparametric magnetic resonance imaging (MRI) for predicting tumor budding (TB) grading in patients with rectal cancer (RC).</p><h3>Methods</h3><p>A retrospective cohort of 458 patients with pathologically confirmed rectal cancer (RC) from three institutions was included. Among them, 355 patients from Center 1 were divided into two groups at a 7:3 ratio: the training cohort (<i>n</i> = 248) and the internal validation cohort (<i>n</i> = 107). An additional 103 patients from two other centers served as the external validation cohort. Deep learning models were constructed for T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) based on the CrossFormer architecture, and deep learning features were extracted. Subsequently, a feature ensemble module based on the attention mechanism of Transformer was used to capture spatial interactions between different imaging sequences, creating a multiparametric ensemble model. The predictive performance of each model was evaluated using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA).</p><h3>Results</h3><p>The deep learning model based on T2WI achieved AUC values of 0.789 (95% CI: 0.680–0.900) and 0.720 (95% CI: 0.591–0.849) in the internal and external validation cohorts, respectively. The deep learning model based on DWI had AUC values of 0.806 (95% CI: 0.705–0.908) and 0.772 (95% CI: 0.657–0.887) in the internal and external validation cohorts, respectively. The multiparametric ensemble model demonstrated superior performance, with AUC values of 0.868 (95% CI: 0.775–0.960) in the internal validation cohort and 0.839 (95% CI: 0.743–0.935) in the external validation cohort. DeLong test showed that the differences in AUC values among the models were not statistically significant in both the internal and external test sets (<i>P</i> &gt; 0.05). The DCA curve demonstrated that within the 10–80% threshold range, the fusion model provided significantly higher clinical net benefit compared to other models.</p><h3>Conclusion</h3><p>Compared to single-sequence deep learning models, the attention mechanism-based multiparametric MRI fusion model enables more effective individualized prediction of TB grading in RC patients. It offers valuable guidance for treatment selection and prognostic evaluation while providing imaging-based support for personalized postoperative follow-up adjustments.</p><h3>Graphical abstract</h3><div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":7126,"journal":{"name":"Abdominal Radiology","volume":"50 10","pages":"4483 - 4494"},"PeriodicalIF":2.2,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143750043","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}