Transfer learning improves AI breast tomosynthesis lesion detection across Western and Eastern datasets

Artificial intelligence (AI) lesion detection on digital breast tomosynthesis transferred across Western and Eastern datasets after fine-tuning: sensitivity 0.90 and intersection over union (IoU) 0.53 Eastern-to-Western; sensitivity 0.89 and IoU 0.56 Western-to-Eastern, both sig…

• Lesion types differed significantly between Western and Eastern databases (P < 0.001), with calcification more prevalent in the Eastern database (60%).
• Models were developed using 94 Western and 157 Eastern patients; without transfer learning, performance dropped significantly when applied across datasets (P < 0.001).

Automated summary

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