Attention-guided CNN-LSTM hybrid model for detecting lung and colon cancer

Abstract

Lung and colon cancer rank most lethal forms of cancer worldwide, particularly in underdeveloped countries where cancer is often perceived as synonymous with death. This view results from difficulties in early detection and restricted access to therapies, which make treating certain illnesses more difficult. To enhance the diagnostic process, a sophisticated deep learning model was developed incorporating an attention mechanism-based hybrid architecture combined with LSTM-based method for identifying colon and lung cancer. With a dataset of 25,000 histopathology pictures, the suggested model detects cancer with an astounding 99.85% accuracy. Comparative analysis were conducted against existing state-of-the-art models, including VGG16, MobileNet, VGG19, DenseNet121, ResNet50, InceptionV3, and an ensemble model, which demonstrated accuracies of 98.65%, 99.05%, 98.45%, 98.45%, 78.90%, 95.80%, and 98.60%, separately. These results highlight the outstanding functionality of the suggested approach in tackling the urgent need for precise and effective cancer diagnosis.