Interpretable Diagnosis of Pulmonary Emphysema on Low-Dose CT Using ResNet Embeddings

TL;DR

This paper introduces an interpretable deep learning method for detecting pulmonary emphysema in low-dose CT scans with high accuracy and low computational cost.

Contribution

The novel contribution is a quality-controlled, interpretable pipeline using ResNet embeddings for emphysema diagnosis without retraining the model.

Findings

The pipeline achieved strong diagnostic performance with ROC-AUC of 0.996 and balanced accuracy of 0.931.

ResNet embeddings combined with QCT markers improved robustness and interpretability of the model.

The method is computationally efficient and suitable for large-scale population screening.

Abstract

Accurate and interpretable detection of pulmonary emphysema on low-dose computed tomography (LDCT) remains a critical challenge for large-scale screening and population health studies. This work proposes a quality-controlled and interpretable deep learning pipeline for emphysema assessment using ResNet-152 embeddings. The pipeline integrates automated lung segmentation, quality-control filtering, and extraction of 2048-dimensional embeddings from mid-lung patches, followed by analysis using logistic regression, LASSO, and recursive feature elimination (RFE). The embeddings are further fused with quantitative CT (QCT) markers, including %LAA, Perc15, and total lung volume (TLV), to enhance robustness and interpretability. Bootstrapped validation demonstrates strong diagnostic performance (ROC-AUC = 0.996, PR-AUC = 0.962, balanced accuracy = 0.931) with low computational cost. The proposed approach shows that ResNet embeddings pretrained on CT data can be effectively reused without retraining for emphysema characterization, providing a reproducible and explainable framework suitable as a research and screening-support framework for population-level LDCT analysis.

Figures