Auditing Sybil: Explaining Deep Lung Cancer Risk Prediction Through Generative Interventional Attributions

Lung cancer remains the leading cause of cancer mortality, driving the development of automated screening tools to alleviate radiologist workload. Standing at the frontier of this effort is Sybil, a deep learning model capable of predicting future risk solely from computed tomography (CT) with high precision. However, despite extensive clinical validation, current assessments rely purely on observational metrics. This correlation-based approach overlooks the model's actual reasoning mechanism, necessitating a shift to causal verification to ensure robust decision-making before clinical deployment. We propose S(H)NAP, a model-agnostic auditing framework that constructs generative interventional attributions validated by expert radiologists. By leveraging realistic 3D diffusion bridge modeling to systematically modify anatomical features, our approach isolates object-specific causal contributions to the risk score. Providing the first interventional audit of Sybil, we demonstrate that while the model often exhibits behavior akin to an expert radiologist, differentiating malignant pulmonary nodules from benign ones, it suffers from critical failure modes, including dangerous sensitivity to clinically unjustified artifacts and a distinct radial bias.

AI models like Sybil show great promise in predicting lung cancer risk from chest scans. However, we do not fully understand how these black-box models make decisions, which is risky for real-world clinical use. They might rely on flawed reasoning rather than actual medical signs. To test the model's reasoning, we built a tool that acts like a digital surgeon. It uses advanced image generation to artificially insert or remove lung nodules (small tissue growths) in patient scans. By feeding these altered scans back into the AI, we can precisely measure how much a specific nodule influences the model's final risk prediction. While the AI often reasons correctly like an expert radiologist, our audit revealed dangerous blind spots. The model sometimes ignores cancers at the lung edges and erroneously flags unrelated artifacts—like ECG electrodes on a patient's skin—as high risk. Uncovering these hidden flaws is a vital step toward making medical AI safe and accountable before hospital deployment.