Evaluation of conventional histopathological scoring in breast carcinoma using artificial intelligence technologies

Abstract

Aim: Breast cancer is the most commonly diagnosed malignancy in women, and early detection plays a critical role in the success of treatment. The Ki-67 proliferation index is widely used to evaluate tumor cell proliferation; however, its manual scoring process is observer-dependent, time-consuming, and inherently subjective. This study aims to assess Ki-67 immunohistochemical staining using deep learning algorithms in an objective, rapid, and reproducible manner, and to compare the model’s performance with conventional scoring methods. 

Materials and Methods: In the first phase of the study, a dataset was created using digital images of Ki-67-stained histological sections obtained from patients diagnosed with breast cancer. These images were used to train a machine learning algorithm. In the second phase, 50 new Ki-67-stained tissue sections previously unseen by the model were digitized, and the model’s predictions were compared with Ki-67 index values calculated by conventional manual assessment.

Results: The developed model achieved a mean absolute error (MAE) of 8.69%, a root mean square error (RMSE) of 13.00%, and a coefficient of determination (R²) of 0.540 in overall prediction performance. For cases with low proliferation (Ki-67<20%), the model demonstrated high accuracy (MAE: 5.31%). Binary classification based on a 20% threshold yielded 80% accuracy, 80% sensitivity, 90% precision, and an F1 score of 0.84.

Conclusion: The use of artificial intelligence algorithms in Ki-67 assessment demonstrated successful performance, with an MAE of 8.69%, and has the potential to reduce pathologists' workload during the preliminary evaluation phase. The findings suggest that, with further refinement, the proposed model could contribute to more objective, consistent, and reproducible assessments in breast cancer diagnostics.