GACNNXAI: Employing Genetic Algorithm-Enhanced Convolutional Neural Networks and Explainable Artificial Intelligence and its Applications
Sakshi Taaresh Khanna, Sunil Kumar Khatri, Neeraj Kumar Sharma · 2024
We have proposed a Genetic Algorithm Enhanced Convolutional Neural Network coupled with Explainable Artificial Intelligence for oral cancer detection in medical imaging. Our approach associated CNN with an explanation method to make CNN-based diagnostic decisions more understandable for a clinician. A high resolution of oral imaging is a challenge due to downsampling in several CNN networks. By optimizing the hyperparameters of CNN using Genetic Algorithms, we compensated for downsampling. Our model's hyperparameters, such as the number of layers, filter size, and learning rate, were adjusted based on performance in the validation set. Advancing further in the interpretability, we have used Layer-wise Relevance Propagation, a Local Interpretable Model-agnostic Explanation technique to indicate CNN's decision-making rationale by selecting influential regions of input images. Our proposed GA-CNN model applied to a comprehensive oral image dataset resulted in a classification accuracy of 93.7%. Our sensitivity and specificity rates were 92.5% and 94.9% improvement over other CNN models available in the literature. We have demonstrated that our GA-based optimization mechanism performs better for parameter searching in the CNN. Our proposed AI tool with an explanation for oral cancer screening will assist clinicians in the early identification of disease, establishing a complementary relationship between human health service providers and machine learning technologies for improved patient outcomes.