Cascaded AI framework matches senior radiologists for ossicular chain malformation detection on CT

A cascaded AI framework diagnosed ossicular chain malformations on CT with sensitivity 0.949 and specificity 0.800, matching senior radiologists and outperforming juniors.

• Retrospective study of 2,462 temporal bone CTs; cascaded nnU-Net segmented auditory ossicles, then discrimination models for each ossicle were integrated.
• Patient-level test-set accuracy 0.874, sensitivity 0.949, specificity 0.800, equalling senior radiologist performance and surpassing junior radiologists and alternative algorithms.
• Multi-site temporal external validation maintained robust ossicle-level AUCs (0.972–0.973) and patient-level accuracy 0.960.

Automated summary

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