What Makes Transfer Learning Work for Medical Images: Feature Reuse & Other Factors

Matsoukas, ChristosKTH,Beräkningsvetenskap och beräkningsteknik (CST),Stockholm, Sweden.;Sci Life Lab, Stockholm, Sweden.;AstraZeneca, Gothenburg, Sweden. (författare)

What Makes Transfer Learning Work for Medical Images : Feature Reuse & Other Factors

Part of proceedings ISBN 978-1-6654-6946-3QC 20230131
Transfer learning is a standard technique to transfer knowledge from one domain to another. For applications in medical imaging, transfer from ImageNet has become the de-facto approach, despite differences in the tasks and image characteristics between the domains. However, it is unclear what factors determine whether - and to what extent transfer learning to the medical domain is useful. The longstanding assumption that features from the source domain get reused has recently been called into question. Through a series of experiments on several medical image benchmark datasets, we explore the relationship between transfer learning, data size, the capacity and inductive bias of the model, as well as the distance between the source and target domain. Our findings suggest that transfer learning is beneficial in most cases, and we characterize the important role feature reuse plays in its success.

Fredin Haslum, JohanKTH,Beräkningsvetenskap och beräkningsteknik (CST),Science for Life Laboratory, SciLifeLab,AstraZeneca, Gothenburg, Sweden.(Swepub:kth)u13aplea (författare)
Sorkhei, MoeinKTH,Intelligenta system,Science for Life Laboratory, SciLifeLab(Swepub:kth)u1h9wx5m (författare)
Soderberg, MagnusAstraZeneca, Gothenburg, Sweden. (författare)
Smith, Kevin,1975-KTH,Science for Life Laboratory, SciLifeLab,Beräkningsvetenskap och beräkningsteknik (CST)(Swepub:kth)u1l33jpf (författare)
KTHBeräkningsvetenskap och beräkningsteknik (CST) (creator_code:org_t)

Ingår i:2022 IEEE/CVF conference on computer vision and pattern recognition (CVPR): Institute of Electrical and Electronics Engineers (IEEE), s. 9215-9224

Av författaren/redakt...: Matsoukas, Chris ...; Fredin Haslum, J ...; Sorkhei, Moein; Soderberg, Magnu ...; Smith, Kevin, 19 ...

Om ämnet

Kungliga biblioteket hanterar dina personuppgifter i enlighet med EU:s dataskyddsförordning (2018), GDPR. Läs mer om hur det funkar här.
Så här hanterar KB dina uppgifter vid användning av denna tjänst.