A survey on Classification of Medical Images using Deep Learning

Abstract

The advancement of technology is creating ease in every field and one of the major fields that has benefitted from this is healthcare. Medical images which includes Multi resonance images, Computed tomography scans and many more have always been used significantly to get a better understanding of the disease, rather knowing the disease and blurred visions of the same can be a matter of concern since so much depends on the images as they are used for analysation. The Covid19 Pandemic hit the population and made things worse for humans and created havoc worldwide, and its early detection using the chest X-rays is significant and the use of technology in this can make it an even better process. So, to classify the chest x-ray images into covid19, viral pneumonia, lung opacity or normal can play a notable role in the area of research. For the indicated purpose, a pretrained model is used by using transfer learning, an efficentnetb4 is used for the same. For the evaluation part of the model, F1 score, sensitivity and precision are used. The method successfully classifies the chest x-ray images into four types, accompanied by an accuracy of 96.73% on the test set. Brain tumor has also been an area of major concern for a long time and various research has been done in the indicated field, but still a lot is left to be discovered. The classification of brain tumors not only helps in creating ease with treatment, but it also depicts that if a tumor is cancerous or not. For this purpose, MRI images are being used for classifying brain tumors into glioma, pituitary, no tumor, meningioma and for this also, a pretrained model is used by using transfer learning, efficentnetb4 is used for the same and for the evaluation part of the model, F1 score, sensitivity and precision are used. The method successfully classifies the brain mri images into four types with accuracy of 98.58% on the test set. While most of the existing techniques either use different methods for same modality or different methods for different modalities, this work uses same method efficientnetb4 for both different types of medical images x-ray as well as mri images.