Abstract

This project aims to develop a robust and efficient method for the detection and segmentation of lung lesions from medical imaging data, particularly focusing on Computed Tomography (CT) scans. Applied deep learning and CNN to detecting and classified lung disease using imagery data. We used the images that belong to 4 categories: healthy, covid-19, viral pneumonia, and bacterial pneumonia. Each category is consisting of 133 images and we used it to develop model that could detect and classify the images in less than 1 minutes. For the image dataset that is used on this project put it on google drive. I put label in each image’s categories with numbers from 0 to 3 (0 = covid-19, 1 = normal, 2 = viral pneumonia, 3 = bacterial pneumonia) 1. Understand the Problem Statement and Business Case Deep learning has been proven to superior in detecting and classifying disease using imagery data. Skin cancer could be detected more accurately by Deep Learning than by dermatologist (2018). Human dermatologist can detect skin cancer with 86.6% accuracy while deep learning can detect skin cancer with 95% accuracy. (Reference: "Computer learns to detect skin cancer more accurately than doctors". The Guardian. 29 May 2018) In this project, we try to develop a model that could detect and classify lung disease using 133 X-Ray images that belong to 4 classes: Healthy, Covid-19, Bacterial Pneumonia, and Viral Pneumonia. To import our data, we used image generator to generate tensor images data and normalize them. We used 428 images for training (80%) and 104 images for our validation (20&). Before we generate our data, we perform shuffling to prevents the model from learning the order of the training. We generate a batch of 40 images and labels. The following is label names for each classification: 0 = Covid- 19 1 = Normal 2 = Viral Pneumonia 3 = Bacterial Pneumonia.