1 September 2023 in Radiology: Artificial Intelligence
To present a deep learning segmentation model that can automatically and robustly segment all major anatomic structures on body CT images. In this retrospective study, 1204 CT examinations (from 2012, 2016, and 2020) were used to segment 104 anatomic structures (27 organs, 59 bones, 10 muscles, and eight vessels) relevant for use cases such as organ volumetry, disease characterization, and surgical or radiation therapy planning. The CT images were randomly sampled from routine clinical studies and thus represent a real-world dataset (different ages, abnormalities, scanners, body parts, sequences, and sites). The authors trained an nnU-Net segmentation algorithm on this dataset and calculated Dice similarity coefficients to evaluate the model’s performance. The trained algorithm was applied to a second dataset of 4004 whole-body CT examinations to investigate age-dependent volume and attenuation changes. The proposed model showed a high Dice score (0.943) on the test set, which included a wide range of clinical data with major abnormalities. The model significantly outperformed another publicly available segmentation model on a separate dataset (Dice score, 0.932 vs 0.871; P < .001). The aging study demonstrated significant correlations between age and volume and mean attenuation for a variety of organ groups (eg, age and aortic volume [rs = 0.64; P < .001]; age and mean attenuation of the autochthonous dorsal musculature [rs = −0.74; P < .001]). The developed model enables robust and accurate segmentation of 104 anatomic structures. The annotated dataset ( https://doi.org/10.5281/zenodo.6802613 ) and toolkit ( https://www.github.com/wasserth/TotalSegmentator ) are publicly available. Keywords: CT, Segmentation, Neural Networks Supplemental material is available for this article. © RSNA, 2023 See also commentary by Sebro and Mongan in this issue.
12 January 2026 in Abdominal Radiology
Purpose To develop and evaluate the diagnostic accuracy of a deep learning algorithm for automated gallstone detection on CT. Methods This retrospective single-center study included randomly selected CT scans from January 2018 to June 2019: 493 gallstone-positive, 470 gallstone-negative, and 180 post-cholecystectomy scans according to the original report (reference standard). Deep learning models were trained on segmented gallbladders (452 scans) and gallstone bounding-box labels (400 scans), randomly selected and manually labeled. All models used 5-fold cross-validation. Gallstone characteristics were collected (e.g., quantity, size, conspicuity). Where available, ultrasound and MRI reports within six months of CT were reviewed for cholecystolithiasis. Diagnostic performance was evaluated on a test set of 90 scans (30 per category). Statistical analyses included Dice coefficient, sensitivity, specificity, and ROC analysis. Results Ninety patients (mean age 64.9 ± 17.9 years; 46 women) were evaluated. The segmentation algorithm achieved a median Dice score of 94.4% (IQR 4.5%). The detection pipeline identified gallstone-positive scans with a sensitivity of 96.7% (95% CI: 90.2–100.0) and specificity of 83.3% (95% CI: 70.0–96.7). Sensitivity was highest for hyperdense (97.3%) and large stones (>10 mm, 97.1%) but lower for gaseous (71.4%), isodense (78.6%), and small stones (<5 mm, 88.9%). Cholecystectomy was correctly identified in 76.7% of cases. Gallstone presence correlated with larger gallbladder volumes (mean 42.8 mL ± 38.3 vs. 24.0 mL ± 20.7, p < 0.001). Ultrasound/MRI comparisons (available for 178 [18.5%] of CTs with a gallbladder) showed low false-positive (0.6%) and false-negative (4.5%) rates for CT. Conclusion The algorithm detects gallstones in CT scans with high sensitivity, particularly for large, conspicuous stones.