Long-axis MRI Segmentation of Hypertrophic Cardiac Myopathy Based on Complete Pseudo Labeling of Mean Teacher

Modern medical imaging technology has advanced quickly, and MRI is now often utilized in clinical settings to help clinicians collect a large number of accurate pictures. For precisely segmenting the heart and identifying cardiomyopathy and other associated illnesses, high-resolution MRI offers the essential circumstances. Semantic segmentation of long-axis, three-chamber cardiac MRI data is a crucial step in the identification of obstructive hypertrophic cardiac myopathy isorders. The popular deep learning segmentation approach, which necessitates a high number of pixel-level annotations in the training process, is challenged by the fact that the acquisition of annotated data in the medical area necessitates expert knowledge and takes a lot of time. To overcome this difficulty, we created a mean instructor model based on complete pseudo labels and a semi-supervised segmentation technique. The teacher network is given noise in this model, which abandons the traditional technique of merely choosing "good"pseudo labels and fully utilizes "bad"pseudo labels. We iteratively train the model until the required performance is reached, using the anticipated entropy to push the "poor"pixels into a sort queue of negative samples. Our semi-supervised segmentation algorithm successfully segments long-axis MRI, boosting segmentation accuracy while lowering labelling costs, according to the experimental results.