High-Precision Lung Cancer Classification with Custom CNN: Evaluation and Transfer Learning for Broader Cancer Types

In the present era, approaches incorporating machine learning have progressed significantly in the field of medical sciences. Globally, Cancer-related deaths are predominantly driven by lung cancer, affecting mortality rates across all demographics. Precise prediction and evaluation of treatment effectiveness are vital for optimizing therapeutic strategies and improving patient outcomes. In our study, we designed an automated study for lung cancer detection using a feature extraction (FE) network. We developed a custom FE model with 15000 lung histopathological images of resolution 200x200 pixels, the performance of which was compared with three pre-defined models—VGG16, VGG19, and Google Net. The crafted model outperformed the pre-trained models achieving an outstanding accuracy rate of 99.87% in classifying lung cancer images. The models were trained for 50 epochs with an update rate of 0.0003. Further, to examine the developed model reliability across various scenarios, we utilized the model’s architecture as a feature extraction layer to develop two additional models for breast cancer and colon cancer classification through a transfer learning approach. An accuracy of 75.43% and 68.14% was obtained by the developed model in classifying colon cancer and breast cancer respectively. The results suggest that the developed model with transfer learning approaches, holds promise for effectively categorizing additional carcinoma types.