Enhancing Accuracy in Ki-67 Scoring: A Correlation-Based Approach Using Automated Image Analysis

The scoring of Ki-67 biomarker expression on immunohistochemistry (IHC) images is an important indicator in breast cancer diagnosis and prognosis. However, the manual scoring process still faces challenges such as subjectivity, interrater variability, and time inefficiency. This study proposes an automated scoring method based on correlation analysis and deep learning to improve scoring accuracy and consistency. The developed pipeline includes image preprocessing (grayscale, thresholding, contour segmentation) and feature extraction of color, texture (GLCM), morphology, and pixel intensity histogram. In addition, spatial and structural features were analyzed using the VGG16 CNN architecture sequentially to detect positive expressions contextually. The correlation between grayscale pixel intensity and scoring values showed that the intensity range of 70-150 was closely related to positive cell areas. Evaluation using regression models showed that the Extra Trees Regressor produced the best performance with an R of 0.902. These results prove that the combination of textural, spatial, and statistical features can improve the effectiveness of the automated scoring system. This method works well despite differences in staining and could be used for other IHC biomarkers in AI supported digital pathology systems.