Efficient Classification of Colon Cancer Histopathology Images Using Hybrid Features and PSO-Optimized Neural Networks

Despite significant advancements in colon cancer detection and classification, several limitations and research gaps persist. A new approach to classifying colon histology images into four groups-normal, malignant well, malignant poor, and malignant moderate-is presented in this work. The proposed approach begins with preprocessing the images to standardize staining variations and illumination differences, ensuring consistent and reliable input for further analysis. Following normalization, key features are extracted using Shi- Tomasi corner detection and Gabor wavelets. Shi-Tomasi features provide precise localization of significant points within tissue structures, while Gabor wavelets capture detailed texture information across multiple scales and orientations. These extracted features form a hybrid feature set that encapsulates both geometric and textural characteristics of the images. This comprehensive feature set is then used as input to a neural network optimized using Particle Swarm Optimization (PSO), enhancing the network's parameter tuning for optimal classification performance. Several histological colon image datasets are used to validate the proposed approach, which shows its superior accuracy and resilience in categorizing various tissue types. This method integrates advanced preprocessing, hybrid feature extraction, and PSO optimization to significantly improve the accuracy and effectiveness of colon cancer diagnosis. It also gives pathologists a useful tool and improves treatment planning and patient outcomes. The method achieves a maximum accuracy of 93.3%.