AI-powered litigation support for radiology malpractice cases. Internal reports and analysis for the Evidentia team.
Comprehensive comparison of Deep Learning vs Classical ML vs Radiomics for predicting nodule subtlety (Ridit score). Includes interactive charts, per-fold results, statistical significance tests, and a future roadmap. Key finding: DL achieves r=0.60, ~3x better than classical approaches.
High-level summary of the proof-of-concept results using the cumulative ordinal model on LIDC-IDRI data. Detection probability predictions and per-threshold accuracy metrics.
Breakthrough finding: the model performs better without radiologist-derived features, using only automated morphological measurements. Demonstrates the system can operate without subjective inputs.
Non-technical explanation of the POC results for stakeholders, legal teams, and business audiences. Translates model metrics into plain language about what the system can and cannot do.
Technical progress report covering the engineering milestones: data preprocessing pipeline, model architecture, training infrastructure, and deployment readiness.