Advancing Lung Disease Diagnosis in 3D CT Scans

2025-07-01

AI Summary

To address non-pulmonary interference, computational redundancy, and severe class imbalance—particularly for rare pulmonary subtypes such as squamous cell carcinoma—in 3D chest CT diagnosis, this paper proposes an end-to-end optimized framework. First, precise lung segmentation removes irrelevant anatomical structures to focus on pathological regions. Second, a lightweight, adapted ResNeSt50 backbone extracts discriminative 3D contextual features. Third, a class-weighted cross-entropy loss is introduced to explicitly mitigate long-tail distribution bias. Evaluated on the Fair Disease Diagnosis Challenge validation set, the model achieves a Macro F1 Score of 0.80—significantly outperforming baseline methods. The framework demonstrates robust diagnostic performance, improved class fairness, and efficient inference, offering a clinically deployable solution with low computational overhead and enhanced interpretability for auxiliary pulmonary diagnosis.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Computer Vision: 3D Computer Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

To enable more accurate diagnosis of lung disease in chest CT scans, we propose a straightforward yet effective model. Firstly, we analyze the characteristics of 3D CT scans and remove non-lung regions, which helps the model focus on lesion-related areas and reduces computational cost. We adopt ResNeSt50 as a strong feature extractor, and use a weighted cross-entropy loss to mitigate class imbalance, especially for the underrepresented squamous cell carcinoma category. Our model achieves a Macro F1 Score of 0.80 on the validation set of the Fair Disease Diagnosis Challenge, demonstrating its strong performance in distinguishing between different lung conditions.

Problem

• Improving lung disease diagnosis accuracy in 3D CT scans
• Reducing computational cost by removing non-lung regions
• Addressing class imbalance in lung disease classification

Innovation

• Remove non-lung regions to focus on lesions
• Use ResNeSt50 for robust feature extraction
• Apply weighted cross-entropy to address class imbalance