AI tool classifies ILD patterns on full HRCT volume and improves resident accuracy by 14.8% in multicenter study

AI accurately classified 3 major interstitial lung disease patterns on HRCT (77.8% accuracy vs thoracic experts’ median 75.6%). For radiology residents, AI assistance improved accuracy by 14.8% and cut reading time by 20.7% (P<0.001), but 6/8 residents still performed worse than…

• Retrospective multicenter study; AI developed on 989 HRCTs, externally tested on 108 HRCTs with 8 residents reading with/without AI.
• AI alone accuracy (77.8%) within range of 7 thoracic radiologists (median 75.6%, range 61.1–81.5).
• Limitation: Inconsistent benefit; 75% of residents performed worse than AI alone, and external test set was relatively small.

RadPigeon summaries are original and for information only. They are not clinical advice.