Distilling CT Foundation Models into Editable Concept Bottlenecks for Lung Nodule Malignancy Prediction

2026-08-07

AI Summary

This work addresses the limited interpretability of existing CT foundation models when directly applied to lung nodule malignancy prediction. The authors propose the first approach to distill representations from CT foundation models (CT-FM and FMCIB) into eight radiologist-defined clinical attributes via concept bottleneck modeling, integrating nodule size to construct an interpretable and editable malignancy risk predictor. Using ridge regression–based concept heads and Bootstrap resampling for evaluation, the method achieves an AUROC of 0.86 on the internal test set and 0.73 on external validation, outperforming embedding-probe baselines. Furthermore, the model supports concept-level interventions, enabling feature-wise contribution decomposition and post-hoc prediction correction, thereby significantly enhancing transparency and clinical controllability.

Abstract

Foundation models provide transferable CT representations, but predictions based directly on these embeddings are difficult to interpret. We developed concept bottleneck models that map two frozen CT foundation-model representations to eight radiologist-defined pulmonary-nodule attributes and predict malignancy from the estimated concepts and nodule size. The models included CT-FM, a whole-CT self-supervised encoder using a 96^3-voxel nodule-centered patch, and FMCIB, a nodule-focused contrastive encoder using a 50-mm crop. Eight ridge-regression concept heads were trained on 2,610 LIDC-IDRI nodules. Malignancy models were trained on LUNA25 and evaluated on a held-out internal test set and the external DLCS cohort. Concept fidelity was assessed using five-fold cross-validated R^2, and malignancy discrimination was assessed using AUROC with 95% confidence intervals estimated by patient-grouped bootstrap resampling. Concept fidelity was modest but higher for FMCIB than CT-FM for subtlety (R2, 0.24 vs. 0.11), spiculation (0.17 vs. 0.08), texture (0.17 vs. 0.07), and lobulation (0.15 vs. 0.05). Internally, the CT-FM and FMCIB concept+size models achieved AUROCs of 0.86 (95% CI, 0.80-0.92) and 0.86 (0.79-0.92), respectively. Externally, AUROCs were 0.72 (0.68-0.75) and 0.73 (0.70-0.76), compared with 0.73 for nodule size alone and 0.60 and 0.67 for the corresponding embedding only probes. Additive predictions could be decomposed into feature-level contributions and modified through controlled concept interventions. Concept bottlenecks provided transparent malignancy predictions with discrimination similar to nodule size alone, while differences in concept fidelity suggest that concept recovery depends on the underlying foundation-model representation.

Problem

• interpretability
• lung nodule malignancy prediction
• concept bottleneck
• foundation models
• CT imaging

Innovation

• Concept Bottleneck Models
• Foundation Model Distillation
• Interpretable AI
• Lung Nodule Malignancy Prediction
• Radiology Concept Embedding