Multifaceted validation of TotalSegmentator performance in abdominal organ segmentation on CT scans integrated in clinical-grade PACS

Background As healthcare data becomes increasingly multimodal, radiologists require automated tools with integrated analysis within Picture Archiving and Communication Systems (PACS). In this context, this study evaluates the clinical-grade PACS integration of the open-source TotalSegmentator for abdominal organ segmentation in Computed Tomography (CT) scans. Methods TotalSegmentator models were integrated into a clinical-grade PACS system. Segmentation performance was quantified using Dice Similarity Coefficient (DSC) and Normalized Surface Dice (NSD) across five diverse datasets (N = 1242). To examine dataset heterogeneity, exploratory data analysis was performed, including per-organ statistics and radiomic features analysis. Grad-CAM and Monte-Carlo dropout were applied to evaluate model interpretability and robustness. Clinical validation was conducted on CT scans (N = 89 organ samples) by a senior and a junior expert, including interobserver variability analysis. Results The model achieved high segmentation accuracy for most organs (DSC −.95 CI: 0.85–0.97), with lower performance on the gallbladder, pancreas, and prostate (DSC −.95 CI: 0.71–0.85). Variability across datasets, as reflected in normalized volume (μ±σ: 0.24 ± 0.18) and intensity (μ±σ: 0.39 ± 0.21), aligned with radiomics findings. Grad-CAM and Monte Carlo results offered insights into model behavior and potential areas for improvement. Clinical evaluation showed that 97% of segmentations required minimal manual corrections. However, 7% of the validated cases showed disagreements exceeding 50% of the performance for gallbladder, pancreas, and prostate. Conclusions TotalSegmentator demonstrates robust and generalizable segmentation for the major abdominal organs with poorer performance for gallbladder, pancreas, and prostate. The high rating (97% scored 3/4) of segmentations by radiologists indicates the advantage of integration into clinical grade PACS towards more efficient automated radiological flow.