Pulmonary Nodule Classification with Fine-Tuned Deep Learning Models

Lung cancer is one of the deadliest cancer types that exists today. At an early stage, deep learning techniques assist radiologists in classifying the malignant pulmonary nodules of lung cancer. This paper studies the various pre-trained deep learning models trained on the IMAGENET dataset and fine-tuned on LIDC-IDRI computed tomography images. Seven pre-trained deep learning architectures, ResNet152, ResNet18, SENet154, ResNext101, SEResNext50, MobileNetV2 and EfficientNetB0 are studied. These architectures are selected based on their specific characteristics. Finally, the fine-tuned EfficientNetB0 shows better performance compared to other models. Classification in medical imaging is challenging due to the invariant background present in images. This study may provide researchers with a base to understand the malignancy classification of lung cancer.