Interpretable Deep Learning Classifier Using Explainable AI for Non-Small Cell Lung Cancer

The primary cause of cancer-related deaths globally is lung cancer, and lowering death rates requires early detection. In pursuit of this, scientists have been exploring deep learning techniques to enhance computer-aided diagnosis using computed tomography (CT) for lung cancer. However, deep convolutional neural networks (DCNNs) still face challenges like overfitting and high variance despite their widespread use. This research proposes an Efficientnet B0 model that utilizes chest CT scan images to precisely classify and detect various forms of Non-Small Cell Lung Cancer (NSCLC) while addressing these challenges. We employed t-distributed Stochastic Neighbor Embedding (t-SNE), a technique for visualizing high-dimensional data, to represent the clustering of different NSCLC forms and normal samples across the training, testing, and validation datasets. This visualization provided insight into data distribution and separability, aiding the interpretability of model predictions. Additionally, we leveraged pre-trained models such as InceptionV3, VGG16, CNN, and EfficientNet B0, the latter known for its scalability and efficiency in deep learning tasks. The performance of our model was evaluated using metrics like accuracy, precision, recall, and loss values. The EfficientNet B0 model achieved a classification accuracy of 93.96%, highlighting its potential for improving early lung cancer detection.