A Performance Comparison of VGG16 for Cancer Cell Classification

Abstract

Cancer of the lungs and brain significantly contributes to global health. Lung cancer is often diagnosed at advanced stages when treatment options are limited. Early diagnosis with sophisticated imaging and AI-driven diagnostics can significantly increase the chances of survival by facilitating prompt treatment. Early tumor detection also improves patient outcomes by enabling more effective treatments that slow tumor development. Yet, it remains infrequent worldwide due to limited screening systems, the time-intensive nature of manual analysis, and complicated techniques like biopsy procedures. Artificial Intelligence (AI-driven CAD) Computer aided diagnostic systems offer a promising solution for tumor detection, reducing diagnostic time and improving accuracy. However, these AI models require large, well-annotated cancer datasets for detailed analysis, detection, and classification. Expanding access to high-quality datasets is essential to enhance AI-driven diagnostics and make early detection more accessible and efficient. Most available datasets are either small or not preprocessed for thorough analysis, which restricts the accuracy and generalizability of machine learning models. To address this issue, work generated 3000 segmented images of lung cancer with only nuclei regions in cellular images and evaluated the Visual Geometry Group 16 (VGG16) model to classify multiorgan cancer datasets. The impact of augmentation on model evaluation is also analyzed. VGG16 shows <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 7. 6 \%, 1 0 0 \%, \text { and }}$</tex> 85.7 % accuracy for original Hematoxylin and Eosin (H&E) images of the lung and brain cancer and segmented images of the lung cancer dataset.