Federated Learning Based Histopathological Image Classification for Oral Squamous Cell Carcinoma

Oral Squamous Cell Carcinoma is one of the most common types of oral cancer which poses a significant threat to global health. Early and accurate diagnosis is pivotal for effective treatment and improved patient outcomes. However, the lack of diverse patient data, which is frequently made worse by data privacy restrictions, makes it difficult to construct reliable diagnostic models. Federated learning emerges as a promising solution to these challenges by enabling collaborative model training while preserving data confidentiality. This study introduces a Federated Learning framework to detect Oral Squamous Cell Carcinoma from histopathological images. Multiple clients train Convolutional Neural Network classification models based on their local data. After that, the local model parameters will be transferred to create an aggregated global model. This approach not only ensures a variety in the model training but also protects patient privacy. We have incorporated both IID and non-IID datasets in our study to examine the influence of data distribution on the global model performance. IID datasets contain identical data distribution across all clients, whereas non-IID datasets present a variation in the amount of data in different clients. This study utilizes the Federated Averaging algorithm which ensures a weighted distribution of parameters across clients. Utilizing this aggregation method, our study achieves 90.18% test accuracy using non-IID datasets.