Multi-threshold segmentation of histopathological colorectal cancer images by an enhanced INFO algorithm

Abstract

Colorectal cancer (CRC) is one of the most prevalent and life-threatening cancers worldwide, where early detection is crucial for improving patient outcomes. Image segmentation plays a fundamental role in medical analysis by enabling precise polyp detection, yet traditional methods face computational challenges in multilevel thresholding, an NP-hard problem that requires efficient optimization. To address this, this study proposes CoINFOSC, a centroid opposition-based weighted mean of vectors (INFO) algorithm enhanced with harmonic oscillation, for robust multi-threshold segmentation of CRC pathology images, enhancing diagnostic precision and efficiency. The proposed method leverages Kapur entropy as the objective function to determine optimal thresholds, enabling precise segmentation of histopathology images. CoINFOSC integrates centroid opposition and harmonic oscillation strategies to improve the exploration–exploitation balance, enhance population diversity, and prevent premature convergence, thereby achieving highly accurate solutions. The optimization performance of CoINFOSC is rigorously validated through extensive experiments on twenty-five unimodal and multimodal benchmark functions and further tested on the IEEE CEC 2017 for dimensions 30 and 50, and on the IEEE CEC 2019 test suites, demonstrating its superior convergence accuracy and robustness compared to state-of-the-art algorithms. Extensive experiments were conducted to evaluate CoINFOSC’s performance, with segmentation results assessed using six metrics: PSNR (27.72862), SSIM (0.81629), FSIM (0.93167), UIQI (0.17803), QILV (0.97781), and HPSI (0.62943). Furthermore, region-based clinical segmentation metrics, including the Dice coefficient and Jaccard index, were evaluated against expert-annotated ground truth masks. The results demonstrate that CoINFOSC outperforms state-of-the-art algorithms across segmentation accuracy, robustness, and convergence speed. The high-quality segmented images produced by CoINFOSC highlight its effectiveness in handling the complexities of CRC pathology images.