Classification of lung cancer with deep learning Res-U-Net and molecular imaging

Lung cancer is a prevalent malignancy, despite the great breakthroughs in detection and prevention, and it remains the important cause of death. In recent days, artificial intelligence has exploded in all fields of science. The use of deep learning in medical science has improved in accuracy and precision of predicting this infestation in the initial stages. In the work, a novel molecular imaging-based Res-U-Net is proposed for classifying two different types of lung cancer. The PET/CT (positron emission tomography/computed tomography) employing an injection F-18-FDG has developed as a useful tool in therapeutic oncologic imaging for both metabolic and anatomic analysis. The proposed model uses Res-U-Net to classify small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC) from normal by using F-18-FDG PET/CT images from the radiogenomics dataset. This dataset images are pre-processed by Gaussian smoothing to reduce the noise from the PET/CT images. Finally, the classification result is obtained through the support vector machine (SVM) classifier which proves the efficiency of the proposed technique. The outcome of the proposed technique yields the best and most accurate results, and it yields the classification accuracy rate of 96.45%for lung cancer into NSCLC and SCLC.