Implementation of Convolutional Neural Networks for Classifying Lung Cancer Types from Histopathological Images
John Patrick Capocyan · 2024
When analyzing histopathological images for lung cancer detection, experts could potentially make mistakes when diagnosing cancer types for their patients. This is significant because misdiagnosing the type of lung cancer would lead to an incorrect treatment costing the patient time, money, and even their life. One potential solution that could solve this problem is the development of a machine-learning model that can accurately distinguish and classify lung cancer types. In this work, I attempt to solve this problem by implementing a convolutional neural network. I trained a DenseNet machine learning model on a dataset consisting of histopathological images of various types of lung cancer. The model reached a 94.07% accuracy with a 0.1646 loss metric in classifying three types of lung cancer. This is significant because the model learned to detect features that are possibly difficult for the human to ascertain. The model also outperforms other state-of-art models that address the same goal. I plan to take this work forward by training and testing additional cancer datasets and potentially training the model to better distinguish between more types of cancer. This research has the potential to save lives if implemented in a clinical setting.