The Effectiveness of Deep Learning Models on the Detection of Pneumonia

Maya Neeranjun

Overview

The purpose of this experiment was to find out if deep learning models accurately diagnose patients with pneumonia. If deep learning models were used with Convolution Neural Network layers and pooling techniques, then the efficiency of detection would increase with max pooling technique and the most amount of CNN layers. If deep learning models were used with Convolution Neural Network layers, then the amount of time the model takes to run would increase with the increase of CNN layers. A model was created with three convolution neural network (CNN) layers and max pooling technique. The number of CNN layers was increased and the pooling technique changed to average pooling technique. The models were trained, validated, and tested using normal and pneumonia x-ray images. Overall, the experiment supported the hypothesis. For both pooling techniques, the increase of CNN layers increased the accuracy and the time it took for the model to run. In addition, the most layers and max pooling led to a higher accuracy of around 74 percent while most layers and average pooling lead to an accuracy of around 49 percent. Different deep learning models using max pooling could be applied to other lung conditions where patients come in with bronchitis or pulmonary edema. Validated models could also be used in other areas of the body to help diagnose patients that have pain.

From the student

At the beginning of the pandemic, I started watching Netflix more often as I could not spend time with my friends. One day, I was wondering how it recommended a similar movie I wanted to watch next, in the “more like this” section. I realized that my computer was using Artificial Intelligence (AI) to predict the next video. Not knowing anything about AI, I started becoming very curious about it which is how I stumbled upon deep learning. During this time, my whole family contracted Covid-19, and my mom was brought into the hospital for developing pneumonia due to the coronavirus. Throughout her stay in the hospital, I was wondering how I could potentially help physicians diagnose patients faster with the number of cases rising. I realized that I could use deep learning to detect pneumonia. From then on, research journey began, and my interest in computer science grew by reading more science journals and watching YouTube videos to learn how to code in python. After knowing the basics of python, I started studying what deep learning models were and how to program them. After a few months, I created a model that could input images of chest x-rays with pneumonia and could output a diagnosis. I learned a lot about deep learning and programming, and I am very excited to share my findings with the AJAS community.