A Simple but Effective Training Process for the Few-shot Prediction Task of Early Rheumatoid Arthritis from MRI
Keywords: Deep learning, few-shot learning, rheumatoid arthritis diagnosis, MRI
TL;DR: A pre- and post-training method for few-shot disease prediction task
Abstract: Predicting rheumatoid arthritis (RA) in an early-stage based on MRI can help initiate timely treatment and therefore halt the progression of the disease and increase the possibility of recovery. Deep learning methods are in general highly suitable for this type of labeling tasks. However, applying this approach to RA detection faces challenges from the lack of a large number of samples, difficulty in distinguishing patterns of RA from imaging artifacts, and a wide anatomical variation, leading to the failure of transfer learning based on pre-trained models. In this paper, a pre- and post-training method for this fewshot task is proposed. Based on the clinical MRI data, this method was validated through cross-validation, achieving a significant improvement in AUC, F1 score, and accuracy to the baseline deep learning models. Since these pre- and post-training strategies are intuitive, effective and easy to implement, they can also contribute to other challenging few-shot medical tasks.
Registration: I acknowledge that acceptance of this work at MIDL requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: novel methodological ideas without extensive validation
Primary Subject Area: Detection and Diagnosis
Secondary Subject Area: Learning with Noisy Labels and Limited Data
Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.