A Fine-Tuned ResNet50 Model for Multiclass Cancer Prediction

Cancer has maintained its position as a leading contributor to mortality across the world. To enhance treatment outcomes and elevate cancer survival rates, which exert significant economic and social repercussions, the development of a precise and early cancer prediction model is required. Equally important is a comprehensive grasp of the biological pathways underpinning cancer development. Over the past two decades, the categorization of diverse cancer types has undergone exploration through both machine learning and deep learning techniques. At the forefront of these efforts, molecular-level diagnostics utilizing microarray technology have the potential to provide accurate, unbiased, and methodical cancer classification. Given that the expression patterns across the entire genome are shaped by thousands of genes, the selective identification of pertinent genes becomes crucial for precise cancer diagnostics, given that not all genes are implicated in cancer processes. In this work, a deep learning-based pre-trained ResNet50 model has been utilized for the prediction of multiple types of cancer. The model has been fine-tuned by varying the number of epochs, optimizers and learning rates. This model has achieved an accuracy of 99% at epoch 50 which is higher in comparison to the existing models.