Efficient Transfer Learning for Cardiac landmark Localization Using Rotational Entropy

Samira Masoudi, Kevin Blansit, Naeim Bahrami, Albert Hsiao

Keywords: cardiac landmark localization, rotational entropy

TL;DR: Introduces a new metric based on rotational entropy as model uncertainty for efficient transfer learning

Abstract: Transfer learning is a common technique to address model generalization among different sources, which requires additional annotated data. Herein, we proposed a novel strategy to select new data to be annotated for transfer learning of a landmark localization model, minimizing the time and effort for annotation and thus model generalization. A CNN model was initially trained using 1.5T images to localize the apex and mitral valve on the long axis cardiac MR images. Model performance on 3T images was reported poor, necessitating transfer learning to 3T images. \textit{Rotational entropy}, was introduced not only as a surrogate of model performance but as a metric which could be used to prioritize the most informative cases for transfer learning.

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: Transfer Learning and Domain Adaptation

Secondary Subject Area: Uncertainty Estimation

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.