Enhancing Colon Cancer Prediction in Histopathology with Integrated Deep Learning Models: A Comparative Study on the LC25000 Dataset
Asma Merabet, Asma Saighi, Mohamed Abderraouf Ferradji, Zakaria Laboudi · 2024
Accurate prediction of colon cancer through histopathology images is crucial for diagnosis and treatment. Despite deep learning models showing promise in this domain, a comprehensive analysis on the LC2500 dataset, especially integrating advanced models, remains limited. This research critically evaluates the predictive performance of CNNs, Inception V3, and ResNet50, and further explores the enhancement of CNN and ResNet50 through integration with Inception V3. The hybridization of these models demonstrates a key advancement, demonstrating that the strategic combination of different architectures can lead to superior predictive capabilities. Our findings reveal significant improvements in prediction accuracy with these integrations, underscoring the capability of Inception V3 to amplify model performance. Specifically, the integration of Inception V3 with CNN yielded an impressive accuracy of 99.27%, while its combination with ResNet50 achieved an accuracy of 99.20% and a perfect recall rate. This study not only offers a detailed comparison of existing deep learning approaches but also demonstrates the substantial advancements in colon cancer prediction achieved by leveraging the strengths of combined architectures, setting new benchmarks for future research.