Abstract The astounding success of machine learning algorithms in medical image analysis in recent years intersects with a time of remarkably increased use of electronic medical records and images for diagnosis. This paper introduces the machine learning algorithms which are applied for medical image analysis, focusing on convolutional neural networks, and emphasizing clinical aspects of the field. The advantage of machine learning in an era of big medical data is that important scaled relationships within the data can be discovered algorithmically without much human interaction.
This paper covers all significant research areas and functions of medical image segmentation, classification, detection. Lastly we discuss about research challenges, newly appearing trends, and future development.
Keywords Medical image analysis, Neural networks, Deep learning.
Machine learning algorithms are within realm of possibility to be pervades deeply in every field of medicine, whether it is drug discovery or tumor grade classification, extraordinarily improving the way of practicing a medicine. The favorable outcome of machine learning algorithms at medical image analysis in recent years comes at an providential time when medical records are increasingly digitalized.
The use of electronic medical records (EMR) has increased from 12.8% to 42.6% amongst office-based radiologist in the US. Medical images is an essential part of a patients EMR and is currently scrutinized by human radiologists, who are limited by speed, energy, and experience. It takes lot of time and financial cost to train them. A delayed diagnosis of disease will harm the patient. Therefore, it is ideal for medical image analysis to be execute in an automated, accurate and efficient way using machine learning algorithm without much human intervention.
Medical image analysis in machine learning is an active field of research, it is because of relatively structured and labelled data, and in this area patients and doctors will first interact artificial intelligence systems which are functioning properly and practical to implement. Image analysis is essential for two reasons, First, in terms of actual patient metrics it works as a litmus test (kind of ) which will determine whether this artificial intelligence systems will actually ameliorate patient results and rate of survival. Another reason is that it contributes in proving ground for interaction of human and artificial intelligent system, how responsive patients will be towards health altering choices being made, or assisted by a non-human actor.
Medical practitioners develop visual representation of internal body organs for clinical analysis and medical mediation. These images dig into distinct features that delineate organ anatomy, i.e. shape and texture, and obtain refined physiological characteristics. The aim of medical image analysis is to use features that have been recognized on the images and classify them into predefined classes. Medical practitioners optically analyze the image and use this data which they got from the radiologists to understand a given medical image. Clinical experience, basic and innate knowledge of the physician, and dexterity play very significant roles in this process. However, given the range of image data which is being propagated nowadays, it is very difficult for medical diagnostics to analyze all the key features conquered by medical images, hence it becomes tough for radiologists and other medical specialists to reach up to final decision. In order to supply the diagnostic decision-making process of physicians semi-automatic and fully-automatic methods can be performed by employing computers and intelligence tools.