Taxlifier: Leveraging Disease Taxonomy for Enhanced Multi-Label Classification in Chest Radiography

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

Taxlifier introduces two novel hierarchical multi-label classification techniques, loss-based and logit-based methods, designed to enhance the classification of thoracic diseases in chest X-ray (CXR) images. These methods address challenges posed by multiple pathologies with overlapping visual characteristics by integrating hierarchical relationships among diseases. Evaluated on three large-scale CXR datasets—CheXpert (224,316 CXRs), PADCHEST (160,000 CXRs), and NIH (112,120 CXRs)—both techniques demonstrated significant performance improvements. The logit-based method achieved a 12% increase in accuracy, 13% in AUC, and 24% in F1 scores, while the loss-based method improved accuracy by 11%, AUC by 10%, and F1 scores by 12% compared to baseline. This integration of domain-specific hierarchical knowledge also provides more interpretable clinical decision support.

Key takeaway

For AI Scientists developing automated diagnosis systems for chest radiography, integrating disease taxonomy via hierarchical multi-label classification is crucial. Your models can achieve significantly higher accuracy, AUC, and F1 scores, as demonstrated by the Taxlifier's 12% accuracy improvement. Consider implementing logit-based or loss-based methods to enhance both performance and the interpretability of your clinical decision support outputs.

Key insights

Leveraging disease taxonomy in hierarchical multi-label classification significantly improves chest X-ray diagnosis accuracy and interpretability.

Principles

• Hierarchical disease relationships enhance classification.
• Domain-specific knowledge improves model interpretability.
• Overlapping pathologies benefit from structured approaches.

Method

Two methods: loss-based integrates hierarchy into optimization; logit-based adjusts predicted probabilities based on parent classes in disease taxonomy.

In practice

• Apply hierarchical classification to medical imaging.
• Use logit-based adjustment for probability refinement.
• Integrate disease taxonomy for clinical decision support.

Topics

• Chest Radiography
• Multi-Label Classification
• Disease Taxonomy
• Hierarchical Classification
• Medical Imaging AI
• Computer-Aided Diagnosis

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• Dual-Modal Lung Cancer AI: Interpretable Radiology and Microscopy with Clinical Risk Integration — Integrating CT radiology and H&E microscopy with clinical data significantly enhances lung cancer diagnosis and interpretability.
• Learning from multiple readings for axial spondyloarthritis classification of the sacroiliac joints — Automated MRI analysis of sacroiliac joints can achieve expert-level axSpA lesion classification and reduce reader variability.
• Leveraging whole slide difficulty in Multiple Instance Learning to improve prostate cancer grading — Whole Slide Difficulty (WSD) derived from pathologist disagreement improves MIL prostate cancer grading.
• Scaling Generative Foundation Models for Chest Radiography with Rectified Flow Transformers — A 1.3B-parameter generative foundation model for chest radiographs achieves expert-indistinguishable synthesis and controllable editing.
• Bridging the interpretability gap for medical artificial intelligence models using class-association manifold learning — CAML enhances medical AI explainability by decoupling decision patterns and visualizing global class-associated knowledge.
• Complementary Roles of Image Classification and Vessel Segmentation in AI-Based Screening for Retinopathy of Prematurity Plus Disease in a Kenyan Preterm Cohort — Image classification and vessel segmentation offer complementary strengths for AI-based ROP Plus disease detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.