Deep Networks for Medical Images Classification, a Comparative Study

Machine learning (ML) and deep learning (DL) have had a significant impact on healthcare, particularly in the field of medical picture categorization, in recent years. These technologies provide a multitude of advantages that enhance the quality of patient care and increase the effectiveness of healthcare services. The primary rationale for employing (ML) and (DL) in medical picture categorization lies in their remarkable capacity to identify illnesses at an early stage. Automated classification systems has the ability to rapidly analyses and interpret pictures, surpassing the speed of human capabilities. This is particularly vital for effectively handling extensive datasets and becomes critical in emergency scenarios where prompt diagnosis is of utmost importance. The objective of this investigation is to categorize medical images into four distinct categories: normal, COVID-19, lung opacity, and viral pneumonia. In this paper three deep learning models were implemented: the classical CNN, AlexNet and GoogleNet. The comparative analysis showed that AlexNet achieved an accuracy of 96.74%, a processing time of 225 seconds, and a loss of 0.1192. The GoogleNet architecture recorded an accuracy of 93.98%, a processing time of 233 seconds, and a loss of 0.1945 while the classical CNN achieved an accuracy of 94.56%, a processing time of 315 seconds, and a loss of 0.1298. Results showed that the AlexNet has a little advantage in regrads to GoogleNet and classical CNN which report very close performances'.