Large Language Models for Automated Synoptic Reports and Resectability Categorization in Pancreatic Cancer.

Background Structured radiology reports for pancreatic ductal adenocarcinoma (PDAC) improve surgical decision-making over free-text reports, but radiologist adoption is variable. Resectability criteria are applied inconsistently. Purpose To evaluate the performance of large language models (LLMs) in automatically creating PDAC synoptic reports from original reports and to explore performance in categorizing tumor resectability. Materials and Methods In this institutional review board-approved retrospective study, 180 consecutive PDAC staging CT reports on patients referred to the authors' European Society for Medical Oncology-designated cancer center from January to December 2018 were included. Reports were reviewed by two radiologists to establish the reference standard for 14 key findings and National Comprehensive Cancer Network (NCCN) resectability category. GPT-3.5 and GPT-4 (accessed September 18-29, 2023) were prompted to create synoptic reports from original reports with the same 14 features, and their performance was evaluated (recall, precision, F1 score). To categorize resectability, three prompting strategies (default knowledge, in-context knowledge, chain-of-thought) were used for both LLMs. Hepatopancreaticobiliary surgeons reviewed original and artificial intelligence (AI)-generated reports to determine resectability, with accuracy and review time compared. The McNemar test, t test, Wilcoxon signed-rank test, and mixed effects logistic regression models were used where appropriate. Results GPT-4 outperformed GPT-3.5 in the creation of synoptic reports (F1 score: 0.997 vs 0.967, respectively). Compared with GPT-3.5, GPT-4 achieved equal or higher F1 scores for all 14 extracted features. GPT-4 had higher precision than GPT-3.5 for extracting superior mesenteric artery involvement (100% vs 88.8%, respectively). For categorizing resectability, GPT-4 outperformed GPT-3.5 for each prompting strategy. For GPT-4, chain-of-thought prompting was most accurate, outperforming in-context knowledge prompting (92% vs 83%, respectively; P = .002), which outperformed the default knowledge strategy (83% vs 67%, P < .001). Surgeons were more accurate in categorizing resectability using AI-generated reports than original reports (83% vs 76%, respectively; P = .03), while spending less time on each report (58%; 95% CI: 0.53, 0.62). Conclusion GPT-4 created near-perfect PDAC synoptic reports from original reports. GPT-4 with chain-of-thought achieved high accuracy in categorizing resectability. Surgeons were more accurate and efficient using AI-generated reports. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Chang in this issue.

Use of GPT-4 With Single-Shot Learning to Identify Incidental Findings in Radiology Reports.

GPT-4 identified incidental adrenal nodules, pancreatic cystic lesions, and vascular calcifications in radiology reports with F1 scores of 1.00, 0.91, and 0.99, respectively. The findings indicate a potential role for large language models to help improve recognition and management of incidental imaging findings and to be applied flexibly in a medical context.

A large normative connectome for exploring the tractographic correlates of focal brain interventions.

Diffusion-weighted MRI (dMRI) is a widely used neuroimaging modality that permits the in vivo exploration of white matter connections in the human brain. Normative structural connectomics - the application of large-scale, group-derived dMRI datasets to out-of-sample cohorts - have increasingly been leveraged to study the network correlates of focal brain interventions, insults, and other regions-of-interest (ROIs). Here, we provide a normative, whole-brain connectome in MNI space that enables researchers to interrogate fiber streamlines that are likely perturbed by given ROIs, even in the absence of subject-specific dMRI data. Assembled from multi-shell dMRI data of 985 healthy Human Connectome Project subjects using generalized Q-sampling imaging and multispectral normalization techniques, this connectome comprises ~12 million unique streamlines, the largest to date. It has already been utilized in at least 18 peer-reviewed publications, most frequently in the context of neuromodulatory interventions like deep brain stimulation and focused ultrasound. Now publicly available, this connectome will constitute a useful tool for understanding the wider impact of focal brain perturbations on white matter architecture going forward.

Lateralized Subthalamic Stimulation for Axial Dysfunction in Parkinson's Disease: Exploratory Outcomes and Open-Label Extension.

BACKGROUND: A randomized trial suggested that reducing left-sided subthalamic stimulation amplitude could improve axial dysfunction. OBJECTIVES: To explore open-label tolerability and associations between trial outcomes and asymmetry data. METHODS: We collected adverse events in trial participants treated with open-label lateralized settings for ≥3 months. We explored associations between trial outcomes, location of stimulation and motor asymmetry. RESULTS: 14/17 participants tolerated unilateral amplitude reduction (left-sided = 10, right-sided = 4). Two hundred eighty-four left-sided and 1113 right-sided stimulated voxels were associated with faster gait velocity, 81 left-sided and 22 right-sided stimulated voxels were associated with slower gait velocity. Amplitude reduction contralateral to shorter step length was associated with 2.4-point reduction in axial MDS-UPDRS. Reduction contralateral to longer step length was associated with 10-point increase in MDS-UPDRS. CONCLUSIONS: Left-sided amplitude reduction is potentially more tolerable than right-sided amplitude reduction. Right-sided more than left-sided stimulation could be associated with faster gait velocity. Shortened step length might reflect contralateral overstimulation.

WarpDrive: Improving spatial normalization using manual refinements.

Spatial normalization-the process of mapping subject brain images to an average template brain-has evolved over the last 20+ years into a reliable method that facilitates the comparison of brain imaging results across patients, centers & modalities. While overall successful, sometimes, this automatic process yields suboptimal results, especially when dealing with brains with extensive neurodegeneration and atrophy patterns, or when high accuracy in specific regions is needed. Here we introduce WarpDrive, a novel tool for manual refinements of image alignment after automated registration. We show that the tool applied in a cohort of patients with Alzheimer's disease who underwent deep brain stimulation surgery helps create more accurate representations of the data as well as meaningful models to explain patient outcomes. The tool is built to handle any type of 3D imaging data, also allowing refinements in high-resolution imaging, including histology and multiple modalities to precisely aggregate multiple data sources together.