Applications of conventional machine learning and deep learning for automation of diagnosis: Case study

Computerized diagnostic methods are commonly used in recent years as initial screening methods for the detection of diseases. Machine learning (ML) and deep learning (DL) are extensively used in the medical field for the investigation of various computer-aided diagnostic applications. Remarkable progress has been made toward the development of such systems in various branches of the medical field by many researchers using the ML approach. In the present work, we compare and contrast the suitability of ML and DL for two different kinds of applications. Suitability of ML and DL methods experimented separately for categorization of white blood cells into the five normal types and abnormal types, and classification of uterine cervix images into cancerous and normal. Both the techniques performed equally well in the case of peripheral blood smear images due to visibly clear features of white blood cells. In the case of uterine cervix images, DL performed better compared to that of ML. This is because of features in the images are not differentiable. From this study, we conclude that conventional ML is useful when features are visible, availability of dataset and resource is a constraint. DL is useful when feature engineering is a difficult task and the availability of the resource is not a constraint.