XrayVision AI

Inspiration

For context, we were participating in a competition called HOSA in the medical innovation category. In this competition, you create an innovation and present it to judges, usually medical professionals. Our inspiration came from a relative who was misdiagnosed, causing a long hospital stay. We found out that misdiagnosis is common and that 795,000 people in the U.S. die or are permanently disabled from it BMJ Quality and Safety. We also learned that radiologists often diagnose patients but face many issues, like a shortage of radiologists, a high error rate, and limited access to remote areas. This research made us realize we could create a solution to improve disease detection and accessibility.

What it does

Our innovation is a website application that identifies whether a patient has a disease or not based on medical imaging such as X-rays. Our innovation goes through a process to maximize accuracy. The process begins with the doctor inputting symptoms that the patient is facing to filter out diseases that the patient does not have. Next, the website will compile a list of possible diseases that the patient has by matching the symptoms of the patient with common symptoms for different diseases. After selecting a disease from the list, the doctor can then upload an image of an X-ray scan or medical imaging required to diagnose the disease. The website will then use image/object detection on the uploaded image to determine whether or not the patient has the disease.

How we built it

We built this project by using a framework called Flask. This framework allowed us to code the front end with HTML and CSS, and use JavaScript and Python for the back end. We also separately trained the AI model and saved it in a keras file. So, to run the model, all we had to do was load the model and predict the image/X-ray scan.

Challenges we ran into

• Finding an image dataset to train the model
• Learning the Flask framework as it was unfamiliar to us
• Learning how to train the model and perfect the accuracy of it
• Connecting the front end with the Python code

Accomplishments that we're proud of

• Being able to create object detection models with consistent 85% or higher accuracy
• Being able to create an innovation that has practical use in the healthcare industry

What we learned

Before starting this project, we didn't know much about machine learning and object detection. Working on it not only helped us explore the field of AI but also boosted our confidence in our ability to tackle significant challenges that many people face. Additionally, we were able to familiarize ourselves with the framework Flask (a framework we were not familiar with)

What's next for XrayVision AI

The next steps for XrayVision AI include expanding the range of diseases it can detect. Currently, it identifies three diseases from X-rays, but we aim to add more diseases in the future. Moreover, we plan to broaden its capabilities beyond X-rays to include CT scans and other types of medical imaging for detecting a wider variety of conditions.

Built With

• css
• flask
• html
• javascript
• keras
• python
• tensorflow