Semi-supervised Active Learning for Left Ventricle Segmentation in Echocardiography

Eman I Alajrami, Nasim DadashiSerej, Jevgeni Jevsikov, Patricia Fernandes, Abas Abdi, Isreal Ufumaka, Darrel P Francis, Massoud Zolgharni

Keywords: Echocardiography, Deep active learning, Semi-supervised active learning

Abstract: Training deep learning models requires large labelled datasets, which are expensive and scarce in medical imaging. This study investigates semi-supervised active learning for left ventricle segmentation in echocardiography, aiming to reduce the need for extensive manual expert annotations. A novel technique for identifying reliable pseudo-labels is proposed. Results show a significant reduction in annotation efforts by up to 93%, achieving 99% of the maximum accuracy using only 7% of labelled data. The study contributes to efficient annotation strategies in medical image segmentation.

Submission Number: 145