EfficientNet-B0 and Hybrid CNN: A Comparative Approach to Lung Cancer Detection

One of the major problems with medical images is identification of the peripheral tumors in the lungs at an early stage because of their subtlety and insufficient characteristics that help distinguish them. The study evaluates one of the deep learning models in comparison with another one EfficientNet-B0 and the new hybrid system that combines ResNet50 with DenseNet121 concerning lung CT scan classification. They analyse the IQ-OTH/NCCD data as it contains 1,190 CT images classified as being normal, benign or malignant. Oversampling was also used as a means to add balance to the data, since the classes were unequal in their number. Transfer learning was employed by both models and the two models were initialized to use large-scale image-data-trained weights to enable feature extraction. The hybrid model uses a friendship of the feature maps between the ResNet50 and DenseNet121, through concatenation, in order to collect more information. According to the results of the experiments, the hybrid approach is more accurate than EfficientNet-B0 which proves the usefulness of combining various convolutional neural networks to enhance lung cancer identification. Keywords- Hybrid CNN architecture, Lung cancer detection, Transfer learning, Computer-aided diagnosis.