This multicenter study aimed to develop a deep learning-based autosegmentation model for pancreatic cancer and surrounding anatomical structures using computed tomography (CT) to enhance surgical planning. We included patients with pancreatic cancer who underwent pancreatic surgery at three tertiary referral hospitals. A hierarchical Swin Transformer V2 model was implemented to segment the pancreas, pancreatic cancers, and peripancreatic structures from preoperative contrast-enhanced CT scans. Data was divided into training and internal validation sets at a 3:1 ratio (from one tertiary institution), with separately prepared external validation set (from two separate institutions). Segmentation performance was quantitatively assessed using the dice similarity coefficient (DSC) and qualitatively evaluated (complete vs partial vs absent). A total of 275 patients (51.6% male, mean age 65.8 ± 9.5 years) were included (176 training group, 59 internal validation group, and 40 external validation group). No significant differences in baseline characteristics were observed between the groups. The model achieved an overall mean DSC of 75.4 ± 6.0 and 75.6 ± 4.8 in the internal and external validation groups, respectively. It showed high accuracy particularly in the pancreas parenchyma (84.8 ± 5.3 and 86.1 ± 4.1) and lower accuracy in pancreatic cancer (57.0 ± 28.7 and 54.5 ± 23.5). The DSC scores for pancreatic cancer tended to increase with larger tumor sizes. Moreover, the qualitative assessments revealed high accuracy in the superior mesenteric artery (complete segmentation, 87.5%-100%), portal and superior mesenteric vein (97.5%-100%), pancreas parenchyma (83.1%-87.5%), but lower accuracy in cancers (62.7%-65.0%). The deep learning-based autosegmentation model for 3D visualization of pancreatic cancer and peripancreatic structures showed robust performance. Further improvement will enhance many promising applications in clinical research.

To develop a deep learning model based on Residual Networks (ResNet) for the automated and accurate identification of contrast phases in abdominal CT images. A dataset of 1175 abdominal contrast-enhanced CT scans was retrospectively collected for the model development, and another independent dataset of 215 scans from five hospitals was collected for external testing. Each contrast phase was independently annotated by two radiologists. A ResNet-based model was developed to automatically classify phases into the early arterial phase (EAP) or late arterial phase (LAP), portal venous phase (PVP), and delayed phase (DP). Strategy A identified EAP or LAP, PVP, and DP in one step. Strategy B used a two-step approach: first classifying images as arterial phase (AP), PVP, and DP, then further classifying AP images into EAP or LAP. Model performance and strategy comparison were evaluated. In the internal test set, the overall accuracy of the two-step strategy was 98.3% (283/288; p < 0.001), significantly higher than that of the one-step strategy (91.7%, 264/288; p < 0.001). In the external test set, the two-step model achieved an overall accuracy of 99.1% (639/645), with sensitivities of 95.1% (EAP), 99.4% (LAP), 99.5% (PVP), and 99.5% (DP). The proposed two-step ResNet-based model provides highly accurate and robust identification of contrast phases in abdominal CT images, outperforming the conventional one-step strategy. Automated and accurate identification of contrast phases in abdominal CT images provides a robust tool for improving image quality control and establishes a strong foundation for AI-driven applications, particularly those leveraging contrast-enhanced abdominal imaging data. Accurate identification of contrast phases is crucial in abdominal CT imaging. The two-step ResNet-based model achieved superior accuracy across internal and external datasets. Automated phase classification strengthens imaging quality control and supports precision AI applications.

Artificial intelligence (AI) algorithms and deep learning continue to change the landscape of radiology. New algorithms promise to enhance diagnostic accuracy, improve workflow efficiency, and automate repetitive tasks. This article provides a narrative review of the FDA-cleared AI algorithms which are commercially available in the United States as of late 2024 and targeted toward assessment of abdominopelvic organs and related diseases, evaluates potential advantages of using AI, and suggests future directions for the field.

Radiology reports are essential for communicating imaging findings to guide diagnosis and treatment. Although most radiology reports are accurate, errors can occur in the final reports due to high workloads, use of dictation software, and human error. Advanced artificial intelligence models, such as GPT-4, show potential as tools to improve report accuracy. This retrospective study evaluated how GPT-4 performed in detecting and correcting errors in finalized radiology reports in real-world settings for abdominopelvic computed tomography (CT) reports. We evaluated finalized CT abdominopelvic reports from a tertiary health system by using GPT-4 with zero-shot learning techniques. Six radiologists each reviewed 100 of their finalized reports (randomly selected), evaluating GPT-4's suggested revisions for agreement, acceptance, and clinical impact. The radiologists' responses were compared by years in practice and sex. GPT-4 identified issues and suggested revisions for 91% of the 600 reports; most revisions addressed grammar (74%). The radiologists agreed with 27% of the revisions and accepted 23%. Most revisions were rated as having no (44%) or low (46%) clinical impact. Potential harm was rare (8%), with only 2 cases of potentially severe harm. Radiologists with less experience (≤ 7 years of practice) were more likely to agree with the revisions suggested by GPT-4 than those with more experience (34% vs. 20%, P = .003) and accepted a greater percentage of the revisions (32% vs. 15%, P = .003). Although GPT-4 showed promise in identifying errors and improving the clarity of finalized radiology reports, most errors were categorized as minor, with no or low clinical impact. Collectively, the radiologists accepted 23% of the suggested revisions in their finalized reports. This study highlights the potential of GPT-4 as a prospective tool for radiology reporting, with further refinement needed for consistent use in clinical practice.

Early-stage cirrhosis frequently presents without symptoms, making timely identification of high-risk patients challenging. We aimed to develop a deep learning-based triple-modal fusion liver cirrhosis network (TMF-LCNet) for the prediction of adverse outcomes, offering a promising tool to enhance early risk assessment and improve clinical management strategies. This retrospective study included 243 patients with early-stage cirrhosis across two centers. Adverse outcomes were defined as the development of severe complications like ascites, hepatic encephalopathy and variceal bleeding. TMF-LCNet was developed by integrating three types of data: non-contrast abdominal CT images, radiomic features extracted from liver and spleen, and clinical text detailing laboratory parameters and adipose tissue composition measurements. TMF-LCNet was compared with conventional methods on the same dataset, and single-modality versions of TMF-LCNet were tested to determine the impact of each data type. Model effectiveness was measured using the area under the receiver operating characteristics curve (AUC) for discrimination, calibration curves for model fit, and decision curve analysis (DCA) for clinical utility. TMF-LCNet demonstrated superior predictive performance compared to conventional image-based, radiomics-based, and multimodal methods, achieving an AUC of 0.797 in the training cohort (n = 184) and 0.747 in the external test cohort (n = 59). Only TMF-LCNet exhibited robust model calibration in both cohorts. Of the three data types, the imaging modality contributed the most, as the image-only version of TMF-LCNet achieved performance closest to the complete version (AUC = 0.723 and 0.716, respectively; p > 0.05). This was followed by the text modality, with radiomics contributing the least, a pattern consistent with the clinical utility trends observed in DCA. TMF-LCNet represents an accurate and robust tool for predicting adverse outcomes in early-stage cirrhosis by integrating multiple data types. It holds potential for early identification of high-risk patients, guiding timely interventions, and ultimately improving patient prognosis.

Medical imaging datasets often contain heterogeneous biases ranging from erroneous labels to inconsistent labeling styles. Such biases can negatively impact deep segmentation networks performance. Yet, the identification and characterization of such biases is a particularly tedious and challenging task. In this paper, we introduce HyperSORT, a framework using a hyper-network predicting UNets' parameters from latent vectors representing both the image and annotation variability. The hyper-network parameters and the latent vector collection corresponding to each data sample from the training set are jointly learned. Hence, instead of optimizing a single neural network to fit a dataset, HyperSORT learns a complex distribution of UNet parameters where low density areas can capture noise-specific patterns while larger modes robustly segment organs in differentiated but meaningful manners. We validate our method on two 3D abdominal CT public datasets: first a synthetically perturbed version of the AMOS dataset, and TotalSegmentator, a large scale dataset containing real unknown biases and errors. Our experiments show that HyperSORT creates a structured mapping of the dataset allowing the identification of relevant systematic biases and erroneous samples. Latent space clusters yield UNet parameters performing the segmentation task in accordance with the underlying learned systematic bias. The code and our analysis of the TotalSegmentator dataset are made available: https://github.com/ImFusionGmbH/HyperSORT

Post-surgical prediction of recurrence or metastasis for primary gastrointestinal stromal tumors (GISTs) remains challenging. We aim to develop individualized clinical follow-up strategies for primary GIST patients, such as shortening follow-up time or extending drug administration based on the clinical deep learning radiomics model (CDLRM). The clinical information on primary GISTs was collected from two independent centers. Postoperative recurrence or metastasis in GIST patients was defined as the endpoint of the study. A total of nine machine learning models were established based on the selected features. The performance of the models was assessed by calculating the area under the curve (AUC). The CDLRM with the best predictive performance was constructed. Decision curve analysis (DCA) and calibration curves were analyzed separately. Ultimately, our model was applied to the high-potential malignant group vs the low-malignant-potential group. The optimal clinical application scenarios of the model were further explored by comparing the DCA performance of the two subgroups. A total of 526 patients, 260 men and 266 women, with a mean age of 62 years, were enrolled in the study. CDLRM performed excellently with AUC values of 0.999, 0.963, and 0.995 for the training, external validation, and aggregated sets, respectively. The calibration curve indicated that CDLRM was in good agreement between predicted and observed probabilities in the validation cohort. The results of DCA's performance in different subgroups show that it was more clinically valuable in populations with high malignant potential. CDLRM could help the development of personalized treatment and improved follow-up of patients with a high probability of recurrence or metastasis in the future. This model utilizes imaging features extracted from CT scans (including radiomic features and deep features) and clinical data to accurately predict postoperative recurrence and metastasis in patients with primary GISTs, which has a certain auxiliary role in clinical decision-making. We developed and validated a model to predict recurrence or metastasis in patients taking oral imatinib after GIST. We demonstrate that CT image features were associated with recurrence or metastases. The model had good predictive performance and clinical benefit.

For magnetic resonance imaging (MRI), a short acquisition time and good image quality are incompatible. Thus, reconstructing time-resolved volumetric MRI (4D-MRI) to delineate and monitor thoracic and upper abdominal tumor movements is a challenge. Existing MRI sequences have limited applicability to 4D-MRI. A method is proposed for reconstructing high-quality personalized enhanced 4D-MR images. Low-quality 4D-MR images are scanned followed by deep learning-based personalization to generate high-quality 4D-MR images. High-speed multiphase 3D fast spoiled gradient recalled echo (FSPGR) sequences were utilized to generate low-quality enhanced free-breathing 4D-MR images and paired low-/high-quality breath-holding 4D-MR images for 58 liver cancer patients. Then, a personalized model guided by the paired breath-holding 4D-MR images was developed for each patient to cope with patient heterogeneity. The 4D-MR images generated by the personalized model were of much higher quality compared with the low-quality 4D-MRI images obtained by conventional scanning as demonstrated by significant improvements in the peak signal-to-noise ratio, structural similarity, normalized root mean square error, and cumulative probability of blur detection. The introduction of individualized information helped the personalized model demonstrate a statistically significant improvement compared to the general model (p < 0.001). The proposed method can be used to quickly reconstruct high-quality 4D-MR images and is potentially applicable to radiotherapy for liver cancer.

Sarcopenia, defined as the progressive loss of skeletal muscle mass and function, has been associated with poor prognosis in patients with pancreatic cancer, particularly those with borderline resectable pancreatic cancer (BRPC). Although body composition can be extracted from routine CT imaging, sarcopenia assessment remains underused in clinical practice. Recent advances in artificial intelligence (AI) offer the potential to automate and standardize this process, but their clinical translation remains limited. This narrative review aims to critically evaluate (1) the clinical impact of CT-defined sarcopenia in BRPC, and (2) the performance and maturity of AI-based methods for automated muscle and fat segmentation on CT images. A dual-axis literature search was conducted to identify clinical studies assessing the prognostic role of sarcopenia in BRPC, and technical studies developing AI-based segmentation models for body composition analysis. Structured data extraction was applied to 13 clinical and 71 technical studies. A PRISMA-inspired flow diagram was included to ensure methodological transparency. Sarcopenia was consistently associated with worse survival and treatment tolerance in BRPC, yet clinical definitions and cut-offs varied widely. AI models-mostly 2D U-Nets trained on L3-level CT slices-achieved high segmentation accuracy (mean DSC >0.93), but external validation and standardization were often lacking. CT-based AI assessment of sarcopenia holds promise for improving patient stratification in BRPC. However, its clinical adoption will require standardization, integration into decision-support frameworks, and prospective validation across diverse populations.

Deep-learning models for prostate cancer detection typically require large datasets, limiting clinical applicability across institutions due to domain shift issues. This study aimed to develop a few-shot learning deep-learning model for prostate cancer detection on multiparametric MRI that requires minimal training data and to compare its diagnostic performance with experienced radiologists. In this retrospective study, we used 99 cases (80 positive, 19 negative) of biopsy-confirmed prostate cancer (2017-2022), with 20 cases for training, 5 for validation, and 74 for testing. A 2D transformer model was trained on T2-weighted, diffusion-weighted, and apparent diffusion coefficient map images. Model predictions were compared with two radiologists using Matthews correlation coefficient (MCC) and F1 score, with 95% confidence intervals (CIs) calculated via bootstrap method. The model achieved an MCC of 0.297 (95% CI: 0.095-0.474) and F1 score of 0.707 (95% CI: 0.598-0.847). Radiologist 1 had an MCC of 0.276 (95% CI: 0.054-0.484) and F1 score of 0.741; Radiologist 2 had an MCC of 0.504 (95% CI: 0.289-0.703) and F1 score of 0.871, showing that the model performance was comparable to Radiologist 1. External validation on the Prostate158 dataset revealed that ImageNet pretraining substantially improved model performance, increasing study-level ROC-AUC from 0.464 to 0.636 and study-level PR-AUC from 0.637 to 0.773 across all architectures. Our findings demonstrate that few-shot deep-learning models can achieve clinically relevant performance when using pretrained transformer architectures, offering a promising approach to address domain shift challenges across institutions.