MEDICAL IMAGE CLASSIFICATION USING ALGORITHM SELECTION

Abstract

Accurate medical diagnosis is a significant part of patient treatment. With the emergence of artificial intelligence, the process of medical diagnosis became easier. The most advanced and state-of-the-art models are based on large datasets, which increases the demand for memory and computational resources. Thus, the automated classification and detection of medical images can be a difficult problem due to small data availability. The issue of data scarcity can be addressed through meta-learning, which is known as "learning-to-learn" concept, that leverages both existing data and accumulated prior knowledge by automatically selecting the machine learning algorithms for unseen tasks. This approach has been widely used in classification tasks. This study propose a different approach of model selection based on priority orders of the algorithms. The method uses shallow classifiers to train different tasks and to select the best algorithm for new tasks. The priority based meta-learning demonstrates the potential to enhance classification performance in a cost-effective manner. The proposed method has outperformed the existing state-of-the-art achieving 100\% accuracy on a test set on meningioma classification and 98.3\% accuracy on test set on an adenocarcinoma classification.