Lung cancer is still one of the main causes of death worldwide due to cancer. This is mainly because most patients are diagnosed at an advanced stage, and lung cancer is very complex to interpret in CT scans. Deep learning techniques have demonstrated potential in automating the process of lung cancer analysis. Most of the current methods are limited to performing binary classification or operating as black-box systems, thus making it difficult for clinicians to trust and apply them. Therefore, it is still a significant issue to have a correct multi-class classification combined with transparent decision-making. Here we present NeuroExplain-Net, an explainability-enabled deep learning framework for multi-class lung cancer screening using CT images. The framework combines hierarchical convolutional feature extraction with an attention-guided feature emphasis mechanism to improve lesion-relevant representations while at the same time reducing non-informative background regions. NeuroExplain-Net is tested on a dataset that contains four classes: adenocarcinoma, large cell carcinoma, squamous cell carcinoma, and normal lung tissue, while the performance is measured against VGG16, ResNet50, DenseNet121, EfficientNet-B3, and Vision Transformer models. Results from the experiments show that NeuroExplain-Net has an overall accuracy of 99.37%. Various explainable artificial intelligence (XAI) methods such as Grad-CAM, SHAP, and Integrated Gradients have been used to demonstrate that the proposed model gives more weight to the clinically relevant areas of malignant cases and shows widespread activation for normal scans.
Patro et al. (Mon,) studied this question.