A Lightweight Deep Learning Model for Automatic Diagnosis of Lung Cancer

According to the statistics provided by the World Health Organization (WHO) 2.21 million new cases of lung cancer were found and 1.80 million death reports worldwide due to lung cancer. Identifying lung cancer at an early stage is the best method for saving the life of any patient. When determined early, cancer can respond to the treatment and medication in the best way and thus result in a greater probability of patient’s survival and thus decreasing morbidity, and will also help in reducing its treatment expenditure. Thus, substantial improvements can be made in the lives of the patients suffering from cancer by detecting and prediction of cancer at early stages thus avoiding delays in providing attention. Lung cancer can be diagnosed using MRI, isotope, X-ray and CT scan images. The focus of the research presented in the paper is to develop, implement and evaluate a novel framework on the basis of Clustering and Classification using a CNN based image processing algorithm which can be applied for the early diagnosis of lung cancer. The proposed applied SMOTE algorithm for balancing an unbalanced dataset before it is given to the model for processing. For the input layer, CT-Scan images from lung nodules are pre-processed to distinctively segment the nodule region of interest (NROI) in accordance with the four radiologists’ comments and markings explaining the coordinates and truthful values.