Polyarchitectural Deep Ensemble with Multimodal Feature Fusion for Histopathological Stratification of Colon and Lung Cancer

Colon and lung cancer represent some of the most intricate and lethal neoplastic diseases globally, significantly raising the likelihood of treatment failure if not identified accurately and promptly. The variations found in conventional medical diagnosis techniques, including their reliance on time, restricted sensitivity, and insufficient accuracy, hinder and prolong the diagnostic process. To address these limitations, the implementation of Artificial Intelligence and Deep Learningbased Convolutional Neural Network (CNN) models has begun to drive a transformative shift in the medical field. Through the utilization of DenseNet121 and MobileNetV2 as distinct CNN models, the current investigation was able to achieve a classification accuracy of 99.44% and 99.92%, respectively. Feature fusion technology is presented as a sophisticated method for attaining precise outcomes. In this process, two fusion models, fused model 1 (EfficientNetV2B0 + ResNet50) and fused model 2 (EfficientNetV2S + ResNet50), are constructed. The classification accuracies of these composite frameworks reached 99.97% and 100%, marking a significant milestone in the diagnosis of colon and lung cancer. By merging numerous critical model features, the suggested feature fusion strategy boosts classification accuracy and reliability. The findings will pave the way for advancements in swift, automated, and precise medical diagnostic techniques, significantly contributing to the design, development, and deployment of reliable and intelligent medical assistive technologies moving forward.