Abstract:
This study investigates the performance of four widely used Convolutional Neural
Networks (CNNs) MobileNetV2, InceptionV3, VGG16 and ResNet50 for disease
classification tasks, leveraging transfer learning techniques. The analysis focuses on
evaluating and comparing the accuracy of these models during both training and testing
phases. MobileNetV2 demonstrated slightly lower performance, achieving 95.26% in
training and 95.00%, InceptionV3, although effective, showed the lowest accuracy among
the models, with training and test accuracies of 93.43% and 93.00%, VGG16 closely
followed, with training and test accuracies of 98.78% and 98.00% and in testing ResNet50
emerged as the top-performing model, achieving the highest training accuracy of 99.24%
and test accuracy of 99.00% respectively. Reflecting its efficiency but limited capacity in
comparison to the top models. respectively. These findings highlight the critical role of
model selection in achieving high-performance disease classification and emphasize the
suitability of ResNet50 and VGG16 for such tasks. Furthermore, the study underscores the
potential of transfer learning to enhance the efficiency of CNNs in medical imaging
applications, while also identifying opportunities for further optimization and fine-tuning
of these architectures to improve their performance in specific use cases.
