Interpretable Deep Learning Model for Classification of Lung Nodules for Early Detection of Lung Cancer

Abstract

Lung cancer is a leading cause of cancer-related deaths worldwide. Early detection through low-dose chest computed tomography (CT) scans is crucial for improving patient survival rates. This study investigates the application of interpretable deep learning models for classifying lung nodules in CT scans as benign or malignant, aiding in the early detection of lung cancer. We leverage 3D Convolutional Neural Networks (CNNs) trained on the LUNA22 ISMI dataset, which has 1,176 lung nodules with a collection of lung nodule annotations from anonymized CT scans. We compared four 3D CNN models with different architectures. One as a baseline model, the second one is based on 3D adaptations of well established architecture known as AlexNet, the third one is our proposed 3D CNN model and the fourth is our proposed 3D CNN with CBAM integration. The CBAM module is inserted after the last convolutional block in a deep learning model. This allows it to leverage the rich feature representations learned by the preceding layers and apply attention mechanisms to enhance the most informative features. Our 3D CNN with CBAM model outperformed the other three with an accuracy of 94.06%, AUC of 98.84% and F1-Score of 95.56%. Our results demonstrate promising performance in binary classification of lung nodules. Considering the criticality of explainability in medical applications, we employed 3D Grad-CAM, a technique for interpreting the predictions of our 3D CNNs. This approach provides visual insights into the regions within lung nodule images that contribute most significantly to the model's classification decision. This interpretability fosters trust and facilitates the adoption of deep learning models in clinical settings. Our findings suggest that interpretable deep learning holds promise for improving the accuracy and efficiency of early lung cancer detection through CT scans. By leveraging interpretable models, we can potentially enhance clinical decision-making and ultimately contribute to improved patient outcomes.