Doctor's Third Eye
Need for the hour to help doctors in the diagnosis of patients with coronavirus based on X-rays of patient's lungs affected with corona and X-rays of patient's healthy lungs(from medical association)
Created on 25th April 2020
•
Doctor's Third Eye
Need for the hour to help doctors in the diagnosis of patients with coronavirus based on X-rays of patient's lungs affected with corona and X-rays of patient's healthy lungs(from medical association)
The problem Doctor's Third Eye solves
Today the whole world is having a panic on Corona Virus and diagnosis of this disease in patients has become a serious problem in all hospitals across the world. So, I have come up with building a CNN with unique modifications in the algorithm using the concepts of neural networks that are trained with the input image sets.The machine analyses and learns from the given image_set to classify the X-rays of patients giving an accuracy of 84-86% validated on test set images. Within 1-2 minutes we will be able to diagnose if a patient is a victim of coronavirus or not. Currently, doctors are diagnosing based on the countable case details of corona victims in their region which makes their diagnosis inefficient. Our model is integrated with the website which is dynamic in nature. Now, as doctors keep on updating X-ray scans of new victims from their hospitals, the centralized model gets all the data and trains itself again and again thus becoming an ideal corona diagnoser. Here, for instance, the machine will be able to learn from the corona patient's lung X-ray in Italy and helps to diagnose in India. Unlike normal doctors, it can analyse thousands of patient details in a few minutes and can speed up the process of finding affected people and isolating them from unaffected. The end-users of this project are the doctors who input patient details on the website and yields its results. For all the patients that the machine identifies with corona, it also predicts the risk factor for the patient more appropriately. Based on the research I was able to identify the vulnerable groups of people who are at higher risk(old people, females, persons with chronic lung diseases etc.) and lower risk(young people, they act as carriers of disease and initial risk for them is minimum that increases with time). Also, the doctor has a feature to find out live outbreak details across the world: visualized and shown in google map for doctors to have an idea most affected regions for help.
Challenges I ran into
The number of standard Xray scan images of patients affected with corona is limited currently due to which the accuracy of the model is limited. So, we made our project dynamic in nature, which means the number of X-rays of victims lungs will keep on increasing as doctors diagnose corona in patients across the world. As soon as the project comes into play, the accuracy metric will increase by helping the machine to learn more and analyse better from more X-rays.
Another problem we faced is time lap between the time taken by our model to run (a few minutes) and time for our website to load (few seconds). After a lot of bug detections, we were able to figure out the problem. The solution we came up is to have a default image say result.png in the directory which will get loaded in the place of result a website will refresh itself at regular interval of time. As soon as our model is executed successfully, the output image will get overlapped in the existing image. Now, it automatically loads into the website.
We were not able to get more data on corona affected patients to find a risk factor by implementing regression.py. This also limits the risk factor value accuracy and precision. The solution of making our website dynamic by connecting it with APIs we will be able to increase accuracy for the risk factor with increasing details of victims.
Importing API's from UNIREST and visualizing the data in API in google charts was a challenging task for our team as we never dealt with it before. We were also running short of time as we are building our project from the scratch meanwhile learning new concepts too. Hopefully, we are about to complete our project successfully after overcoming many other challenges too.