Pulmonary nodule decision support system with deep learning approach

Today, the use of machine learning is becoming widespread to facilitate medical analysis and diagnosis. The machines are gained the ability to make comments by developing computer-aided diagnostic systems with machine learning. Thus, computer-aided diagnostic systems increase accuracy of diagnostic and detection by offering a second opinion to radiologists. Early diagnosis increases the patient's survival. In this thesis, a decision support system was developed for the classification of pulmonary nodules (benign or malignant) in the diagnosis of lung cancer using convolutional neural networks which is a method of deep learning approach. In the machine learning, the convolutional neural network (CNN) is a deep artificial neural network that is successfully applied in image analysis. The CNNs use the multi-layering structure which requires minimal pre-processing. 1070 Thorax CT images of 600 patients from the Farabi Hospital Radiology Department of Karadeniz Technical University were used as the data set. The pulmonary nodules were automatically classified for diagnosis of lung cancer from the thorax CT images by using the developed decision support system. As a result of five-fold cross-validation, the highest accuracy was 84%, the average accuracy was 76%, the average sensitivity was 77%, and the average specificity was 75%. The improved system has the highest accuracy among the studies using the basic CNN model.