Automatic disease detection from medical images with data augmentation and deep learning techniques

In 2019, the spread of the Covid-19 disease worldwide negatively affected life. Covid-19 disease is a type of pneumonia disease. Early diagnosis of pneumonia patients with deep learning method is the main subject of this thesis study. Chest X-rays (X-Ray images) of individuals for early diagnosis were used as dataset in this thesis study. This study used coronaHack-Chest-Xray on the Kaggle data site as a dataset. The images in the dataset; It is labeled into two different classes, pneumonia and normal. VGG-16, VGG-19, MobileNet, InceptionV3, and Xception; deep learning models were used separately in this thesis. Each model was run with 4 different methods. Methods; 1- No data augmentation and cross-validation, 2- No data augmentation, cross-validation, 3-Data augmentation, no cross-validation, 4-Data augmentation, cross-validation. When all model and method matching combinations were run, the MobileNet model gave the best success metric results on the CoronaHack-ChestXray dataset, with the no data augmentation-no cross-validation method, training set 80% – test set 20%. The model with the best result, The success metrics of separating the images into 2 classes, pneumonia and normal, are summarized as follows: the accuracy metric is 99%, the precision metric is 98%, and the recall metric is 98%. Other success metric results, complexity matrix, and running success metrics of other model-method combinations were obtained in this thesis. KEYWORDS: Deep Learning, Data Augmentation, Medical Imaging, Covid-19, Pneumonia