Content-based Image Retrieval by Using Deep Learning for Interstitial Lung Disease Diagnosis with Chest CT

Jooae Choe et al. developed a deep learning-powered content-based image retrieval (CBIR) system for diagnosing interstitial lung disease (ILD) using chest CT images. The CBIR system segments lung regions, quantifies six disease patterns (e.g., ground-glass opacity, consolidation), and retrieves three similar cases based on Euclidean distance of feature vectors. The study evaluated the system's effectiveness in improving diagnostic accuracy and interreader agreement among eight physicians with varying experience levels. Using 288 confirmed ILD cases, diagnostic accuracy increased from 46.1% to 60.9% after CBIR assistance, with improvements notable in usual interstitial pneumonia (UIP) and nonspecific interstitial pneumonia (NSIP) cases. Interreader agreement also improved (Fleiss k: 0.32 to 0.47). While CBIR demonstrated limitations for less common patterns (e.g., cryptogenic organizing pneumonia), its ability to support radiologists by offering visually comparable cases has significant implications for ILD diagnosis. The findings emphasize the potential of CBIR for aiding diagnostic consistency in clinical settings. Coreline Soft also participated in this research using aview.