Enhancing Multi-Cancer Classification with VGG and EfficientNet: Evaluating CNN Performance for Automated Detection in Medical Imaging

In this research, an automated, deep learning–based framework for multi-cancer classification was successfully developed, effectively replacing subjective, labor-intensive diagnostic interpretation with a robust and scalable CNN-driven pipeline. Leveraging a transfer learning approach built on VGG16 and EfficientNet-B3 architectures, the system incorporates comprehensive preprocessing, targeted data augmentation, stratified train–test splits, and carefully staged fine-tuning to learn discriminative, multi-scale features that capture subtle tissue morphology, nuclear variations, lesion boundaries, and color-texture distortions across diverse cancer types. Evaluation through accuracy, sensitivity, specificity, confusion matrices, ROC curves, and class-wise probability distributions demonstrated high performance and reliable generalization, while probability-based outputs enhanced interpretability and clinical decision support.   A deployable web application was implemented, featuring an intuitive interface for uploading medical images, obtaining real-time cancer predictions, viewing probability scores, and accessing diagnostic insights. Automated logging, dashboard analytics, and prediction history enable traceability, continuous monitoring, and operational quality improvement. Compared with classical machine learning pipelines that rely on handcrafted features, the proposed deep learning approach significantly reduces manual feature engineering effort, scales effectively to multiple cancer categories, and maintains resilience across varying imaging modalities, acquisition Future work can further enhance the framework at data, model, and deployment levels. Potential improvements include expanding datasets with larger, more diverse clinical samples; incorporating region-based annotations for lesion localization; and integrating stain-normalization pipelines for improved cross-institution consistency. Advanced modeling strategies such as lightweight EfficientNet variants, MobileNet, ConvNeXt, and Vision Transformer (ViT) hybrids may improve efficiency and discriminative power. Additional enhancements may include multi-task learning for combined classification and segmentation, model pruning, quantization, and knowledge distillation for edge-device deployment in low-resource clinical environments. Functional extensions could incorporate heatmap-based explainability overlays (Grad-CAM), uncertainty estimation, IoT-enabled pathology workflows, multimodal fusion with genomic or clinical data, 3D image processing for volumetric scans, and federated learning to enable cross-hospital model improvement without centralizing sensitive patient data. These future directions aim to evolve the framework toward a fully integrated, intelligent, and trustworthy AI system for real-world cancer diagnosis and clinical decision support.

[1]  K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” International Conference on Learning Representations (ICLR), 2015.  [2]  M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” International Conference on  Machine      Learning (ICML), 2019. [3]    A. Esteva, B. Kuprel, R. Novoa, et al., “Dermatologist- level classification of skin cancer with deep neural networks,” Nature, 2017.  [4]   X. Wang, Y. Peng, L. Lu, et al., “ChestX-ray8: Hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases,” IEEE CVPR, 2017.  [5]       J. Cheng, Q. Liu, H. Zhao, “Deep learning for histopathology image analysis: A comprehensive survey,” Medical Image Analysis, 2021.  [6]  S. Sudharshan, R. Petitjean, F. Spanhol, et al., “Multiple-instance learning for histopathological breast cancer image classification,” IEEE Transactions on Biomedical Engineering, 2019.  [7]  H. Fu, J. Cheng, Y. Wong, et al., “Joint optic disc and cup segmentation based on multi-label deep network learning,” IEEE Transactions on Medical Imaging, 2018.  [8]    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.  [9]  J. C. Dunn, A. S. Rallapalli, “Applications of deep learning in cancer imaging: A review,” Journal of Medical Imaging, 2023.  [10]      A. N. Author, “State-of-the-art CNN-based cancer detection and classification using transfer learning,” IEEE Access, 2024.