Histopathological Image Processing for Lung Carcinoma Classification: A Comparative Study of Stain Normalization Methods for CNN-Based Analysis

In the present era, Cancer-related deaths are predominantly driven by lung cancer globally, causing significant deaths across all demographics. Precise prediction and evaluation of treatment effectiveness are crucial for optimizing therapeutic strategies and improving patient outcomes. Approaches leveraging machine learning have made significant strides in medical sciences. In the proposed study, a Convolutional Neural Network (CNN) framework is adopted to develop a machine learning model, trained on a dataset received from ICMR Funded Project cohorts comprising 83 images of lung carcinoma. Additionally, three distinct models were developed using datasets processed with Macenko normalization and histogram equalization techniques, alongside the real dataset. These models aim to enhance classification accuracy by addressing challenges such as stain variation and inconsistent illumination in histopathological images. The best accuracy of $99.45 \%$ was achieved by the model trained on a Histogram equalized dataset, underscoring the potential of integrating image preprocessing techniques with CNN architectures for lung cancer diagnosis.