Protein Ki67 is a protein that serves as a marker for cellular proliferation in cancer research and molecular biology. To determine the expression of Ki67 protein, the Immunohistochemistry (IHC) is used as the most popular technique, in which the determination results are shown using distinct colors. However, effectively detecting the Ki67 protein is a tough task due to the requirement of specific knowledge and significant expertise. Hence, scientists have been exploring alternate approaches, such as machine learning and deep learning, to tackle this problem. The emergence of deep neural networks (DNN) has led to substantial progress in computer vision and image processing, obviating the requirement for manual feature selection. Despite the effectiveness of the deep neural network PathoNet in recognizing Ki67 cells, the reliance on predefined thresholds presents constraints in terms of flexibility and adaptability when applied to diverse datasets. Consequently, this work presents a method for automatically optimizing the threshold in PathoNet, with the goal of enhancing the model’s adaptability and precision. The F1 score in identifying and classifying Ki67 cells was improved by our proposed method, with an increase of 0.3–0.8% compared to existing methods. Our proposed method increased the F1 score in detecting and classifying Ki67 cells by 0.3–0.8% compared to other methods, providing a significant advantage over the current state-of-the-art solutions.