Deep Learning in Oncology: Improving Early Cancer Detection with UNet, EfficientNet, and DenseNet

Medical imaging analysis has been revolutionized by deep learning (DL) techniques, which automate and generate highly accurate diagnostic results. The study examines the use of medical imaging data to evaluate DL pipelines for oncological disease classification. The proposed workflow integrates UNet for segmentation, EfficientNet for feature extraction, and DenseNet for classification, which yields the highest accuracy (∼96.2%) with better precision, recall, and F1-scores. In comparison to other model combinations, such as ResNet and VGG-based classifiers, DenseNet outperforms both because of its dense connection and feature reuse capabilities. EfficientNet has excellent feature extraction performance, effectively capturing hierarchical visual details and improving classification accuracy across various architectures. The findings indicate that utilizing EfficientNet for feature extraction and DenseNet for classification is the best approach for early cancer detection using medical imaging.