DEEP LEARNING MODELING TO CONTRIBUTE IN THE PROCESSING OF MEDICAL IMAGING, APPLICATION FOR BINARY, MULTI-CLASS CLASSIFICATION OF A BRAIN TUMOR

Computer vision, image processing

Computer vision is one of the fastest growing fields thanks to deep learning and the sheer numbers of data circulating today. There are many areas of application, including autonomous car driving, image classification, facial recognition, art, environmental and health domains. The computer vision algorithm uses image segmentation to know each element that makes up the image by combining the image classification algorithm with object localisation and other algorithms.Among the fields mentioned, medical imaging takes a major place in terms of computer vision research. The brain is the main organ, the center of motor activity in the human body. The diagnosis of the brain is very delicate and complex and is the subject of much research and study. Several methods such as MRI, CT scan. In the clinical diagnosis and treatment of brain tumours, the manual reading of images consumes a lot of energy and time, as the acquisition has to be repeated as many times as there are slices, which leads to patient fatigue. MRI takes 30 minutes to 1 hour, generating many unnecessary images which slows down the treatment and makes the patients tired. One solution to help with this technique is the use of a deep learning model that will train multiple medical images, fill in missing data from MRI or CT scans and test the image from an MRI or CT scan to help doctors make a diagnosis. The objective of the present work is to create a robust deep learning algorithm to assist in the binary and multi-class classification of a brain tumour. The algorithm is created in its entirety from scratch. All regularisation techniques to remove overfitting and optimisation techniques to find parameters quickly have been tested and implemented to have a robust algorithm. For the dataset, our method can achieve a maximum accuracy for validation of 96.88% for binary classification, 93.38% for multi-class classification and 98.37% for binary classification using the transfer learning technique.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.