Medical image classification based on curriculam learning

Abstract

With the emergence of large medical images and exceptional growth of diagnostic methods, categorizing them into respective class has always been a dominant topic in computer vision. Though the system seems ubiquitous, achieving higher accuracy rates for classification is critical. Semi-Supervised Learning (SSL) is better than supervised learning as it eliminates labeling all images thus reducing computational cost and time. Existing methods suffer from classification accuracy due to the presence of outliers in critical images. This paper is an attempt to apply SSL through Multi-Modal Curriculum Learning (MMCL) strategy over medical images. Through this, medical images can be categorized into normal and abnormal images. Experimental results demonstrate good accuracy for classification.