TotalSegmentator MRI is an open-source, user-friendly tool that provides automatic, robust sequence-independent segmentations of major anatomic structures on MRI scans; the ready-to-use online tool is available at https://totalsegmentator.com .

This study aims to develop a tool based on deep learning algorithms for automatic liver segment and liver lesion segmentation on Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). We demonstrate its clinical utility using a qualitative example of hepatocellular carcinoma (HCC) response to transarterial chemoembolization (TACE). The models are provided as open-source software to update and improve the capabilities of TotalSegmentator. Liver segmentation was performed on 193 CTs and 120 MRIs, using fivefold cross-validation for training/testing. 429 CTs and 321 MRIs with liver lesions and 15 CTs and 13 MRIs without liver lesions were collected. Of these 414 CTs and 308 MRIs were manually segmented and a nnU-Net was trained on 750 images and tested on 80. Inter-rater variability was examined on 20 CTs and 20 MRIs by two independent readers. We analyzed its potential clinical utility on 172 TACE-treated HCC on CT. Performance was evaluated using sensitivity, false positives, and volume. Voxel-wise segmentation was evaluated using the Dice coefficient. Our model’s liver segmentation achieved Dice coefficients of 0.897 for CT and 0.847 for MRI. Liver lesion detection on CT achieved 75.8% sensitivity, 0.522 false positives per case (FP/c), and 0.658 Dice; on MRI, 62.7% sensitivity, 1.029 FP/c, and 0.337 Dice. Following TACE, median HCC attenuation significantly decreased from 51.33 HU to 38.5 HU. Human readers showed higher agreement (sensitivity: 64.7%, Dice: 0.464) than Lesion model (LM)-reader comparisons (sensitivity: 53.2%, Dice: 0.432) and the LM had a slightly higher FP/c (0.825 vs. 0.775). Overall, our algorithms reliably detect and segment liver segments and lesions on both CT and MRI and the qualitative assessment of HCC response to TACE illustrates the model’s potential value for clinical and research applications.

The aim of this study was to develop an open-source nnU-Net-based AI model for combined detection and segmentation of unruptured intracranial aneurysms (UICA) in 3D TOF-MRI and compare models trained on datasets with aneurysm-like differential diagnoses. This retrospective study (2020–2023) included 385 anonymized 3D TOF-MRI images from 345 patients (mean age 59 years, 60% female) at multiple centers plus 113 subjects from the ADAM challenge. Images featured untreated or possible UICA and differential diagnoses. Four distinct training datasets were created, and the nnU-Net framework was used for model development. Performance was assessed on a separate test set using sensitivity and false positive (FP)/case rate for detection and DICE score and NSD (normalized surface distance, 0.5 mm threshold) for segmentation. Segmentation performance on the test set was also compared to a second human reader. The four models achieved overall sensitivity between 82 and 85% and an FP/case rate of 0.20 to 0.31, with no significant differences ( p = 0.90 and p = 0.16) between them. The primary model showed 85% sensitivity and 0.23 FP/case rate, outperforming the ADAM-challenge winner (61%) and a nnU-Net trained on ADAM data (51%) in sensitivity ( p < 0.05). Mean DICE (0.73) and NSD (0.84 for 0.5 mm threshold) for correctly detected UICA did not significantly differ from human reader performance. Our open-source, nnU-Net-based AI model (available at https://zenodo.org/records/13386859 ) demonstrates high sensitivity, low FP rates, and consistent segmentation accuracy for UICA detection and segmentation in 3D TOF-MRI, suggesting its potential to improve clinical diagnosis and monitoring of UICA.

Purpose: In outpatient settings, extensive patient records must frequently be reviewed under time constraints, making efficient extraction and summarization of key clinical information essential. Large language models (LLMs) are potentially useful for this task but require validation for clinical reliability. This study assesses OpenAI’s GPT-4o for generating structured summaries to assist in neurovascular consultation preparation, aiming to increase efficiency by automating critical data extraction. Methods: A prospective study was conducted from May to August 2024 at a tertiary care hospital, involving a total of 70 patients. Structured summaries were generated by GPT-4o using a predefined template. Extracted data were categorized into aneurysm-specific details, imaging summaries, and patient-specific clinical factors. Accuracy and completeness were assessed by clinicians, with performance measured using precision, recall, specificity, and accuracy. Results: High accuracy (≥ 0.96) was measured across most categories. In aneurysm-and patient-specific data, extraction performance varied based on stability over time. Aneurysm location and other stable details were extracted consistently, while changes in aneurysm size and medication lists showed variations. In rare cases, aneurysm details were misattributed to a different aneurysm within the same patient. Imaging summaries were generally concise and clinically useful, though their effectiveness declined when summarizing multiple prior studies. Conclusion: Neurovascular patient data was effectively structured by GPT-4o, demonstrating high accuracy with minimal errors. While occasional misattributions like outdated information were observed, reliable citation of sources facilitated easy verification. These findings support integrating LLM-generated summaries into neurovascular consultations, with further optimization needed for temporal data tracking and on-premise implementation to address privacy concerns.

Purpose In idiopathic normal pressure hydrocephalus (iNPH), longitudinal monitoring of ventricular size is important for diagnosis and treatment follow-up. This study aimed to validate a fully automated AI model for CT ventricular volumetry with subcompartments and to compare AI-derived volume changes with routine radiology assessments. Methods This retrospective, single-center study included 88 patients with iNPH and 456 non-contrast-enhanced head CT examinations. The model was trained on 38 manually labeled CT scans with 12 ventricular subcompartments. Outcomes included segmentation accuracy, correspondence between AI-derived longitudinal ventricular volume changes and radiology report categories (decreased, unchanged, increased), radiologist detection thresholds for ventricular change, and paired pre- and postoperative volume changes in 22 patients with ventriculoperitoneal shunt. Results Mean segmentation accuracy was high (Dice, 0.83). 91% of 100 segmentations were rated as excellent by an expert neuroradiologist. AI-derived ventricular volume changes corresponded well to radiology report categories (median total ventricular volume changes of -17% in cases reported as decreased, 0% in unchanged cases, and +22% in increased cases; all p < 0.001). Radiologists reported ventricular volume change in 50% of cases at an AI-measured relative volume change of +/-6%, and in 90% of cases at +21% for enlargement and -18% for decrease. After shunt placement, ventricular volume decreased by -8% (median), with the largest relative reductions observed in the right temporal and occipital horns. Conclusions Automated AI-based ventricular segmentation on CT enables accurate and reproducible assessment of ventricular volume changes in iNPH and complements routine radiological evaluation for longitudinal and postoperative monitoring.