Marauder's Lab in 'From Chat Window to Custom Radiology Applications - Creating Tailored Tools with Codex and Claude Code'

Résumé fourni par la source

Marauder's Lab Volume Tool and Dual Lesion Workstation: Two Complementary Python-Based Applications for Advanced CT Lesion Analysis The two software tools described were developed with close help of Large Language Models (ClaudeAI and ChatGPT) with the goal of integrating AI-driven image analysis into routine radiological workflows, while remaining operable on standard clinical workstations without dependency on dedicated GPU hardware. The integrated tools have not been validated for routine use, and their functions remain experimental and unverified. Therefore, they are intended for research purposes only and should be used at the user’s own responsibility. Marauder's Lab Volume Tool (MaraudersLab_pro.exe) is a tkinter-based Windows 11 application for volumetric analysis and segmentation of medical imaging data. It integrates TotalSegmentator (nnU-Net–based) for automated whole-body segmentation from CT scans in DICOM or NIfTI format and operates with CPU-only inference for broad workstation compatibility. Additional modules support detection of pulmonary embolism, lung nodules, kidney cysts, brain structures (SynthSeg), and white matter hyperintensities. The tool also includes an interactive ROI segmentation interface, NRRD export, 2D/3D NIfTI viewers with annotation support, and a multiplanar reconstruction viewer via a PyQt5/VTK subprocess. A self-contained runtime manages temporary data and nnU-Net configuration to ensure stable operation in restricted clinical environments. Beyond interactive use, the application offers a fully automated headless batch mode for unattended processing. When enabled it runs in the background with system tray access. An input folder watcher monitors predefined pipeline-specific directories, detects newly added DICOM datasets after file stability checks, and automatically selects the largest series for processing. A job manager then executes each case sequentially, including data import, segmentation and detection, result aggregation into CSV, structured PDF report generation, and export to organized output folders. The Dual Lesion Workstation (DualLesionWorkstation.exe) is a PyQt5-based application for comparative volumetric assessment of lesions across two time-separated imaging series, typically current and prior CT scans. It displays both image stacks side by side, supports DICOM and NIfTI inputs, and ensures accurate volume calculations via automatic voxel spacing extraction. Segmentation uses semi-automatic intensity-based region growing with robust statistics (median and MAD), followed by Chan–Vese refinement for precise boundaries. Multiple interaction modes enable flexible seeding and ROI definition, including 3D bounded segmentation. The software supports multi-lesion tracking with real-time volume reporting, adjustable sensitivity, and a removal mode for corrections. Standard radiological navigation controls are integrated throughout. Together, these tools form a coherent software ecosystem: the Volume Tool addresses automated, AI-driven detection and volumetric quantification of pathological findings across multiple organ systems, while the Dual Lesion Workstation provides the radiologist with a precision instrument for interactive, observer-guided comparative measurement — complementary approaches that jointly support both screening-oriented discovery and longitudinal response assessment in clinical CT reporting.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.