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

2026-01-30

AI Summary

Although the existing lung cancer risk prediction model Sybil demonstrates strong performance, its decisions rely on correlations rather than causal mechanisms, raising concerns about clinical reliability. This work proposes the S(H)NAP framework—the first causal audit of Sybil—employing a model-agnostic generative intervention attribution method coupled with a 3D diffusion bridge model to precisely manipulate anatomical features in lung imaging, validated by radiologists. The study confirms that Sybil achieves expert-level capability in distinguishing benign from malignant nodules, while also uncovering critical limitations, including sensitivity to non-clinical artifacts and radial bias. These findings establish a new paradigm for enhancing the interpretability and safety of clinical AI systems through causal auditing.

Technology Category

• Reasoning under Uncertainty: Causality
• Computer Vision: Interpretability, Explainability, and Transparency
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

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.

Problem

• causal verification
• deep learning audit
• lung cancer risk prediction
• model interpretability
• Sybil

Innovation

• generative interventional attribution
• causal auditing
• 3D diffusion bridge
• model-agnostic explanation
• Sybil model