Deep Learning and Texture Analysis for Lung and Colon Cancer Predicting

Abstract

Cancer remains a major cause of death worldwide, with lung and colon (LC) cancers presenting significant challenges to healthcare systems due to their high rates of occurrence and mortality. Early and precise diagnosis is essential for better patient outcomes. This research utilizes recent advances in deep learning (DL) and texture analysis (TA) to create a reliable predictive model for detecting LC cancer through histopathological images (HPI). A hybrid method is proposed that combines a gray-level co-occurrence matrix (GLCM) for extracting texture features with an adaptive modified EfficientNet B2 model (AM-EfficientNet B2) for deep feature extraction. These features are used to train custom artificial neural network (ANN) classifiers on the LC25000 dataset. The AM-EfficientNet B2-based model achieves an outstanding overall accuracy of 0.9996 and 1.0 for Precision, F1 score, and Recall, and recorded 0.99975 for AUC, outperforming the GLCM model with an accuracy of 0.9228 and 0.9257, 0.9228, 0.9219, and 0.9517 for Precision, Recall, F1score, and AUC, respectively. The Grad-CAM (Gradient-weighted Class Activation Mapping) method was employed to visually explain the AM-EfficientNet B2 model results, thereby improving transparency and trustworthiness. These results suggest that the proposed DL model has the potential to serve as a valuable tool in assisting clinicians with accurate and reliable cancer diagnoses using HPI, ultimately improving patient outcomes. The study contributes to the expanding research on DL in cancer prediction, highlighting its potential to revolutionize oncology.