Team members - Shubh Singh Garg, Arshdeep Singh
As we all know, Covid-19 Virus was the biggest news of 2020, and as such looking for a problem to solve was not that hard. One of the main headlines and problems, related to covid was the amount of burden it put on the current health system. Be it the large population that needed regular testing, or the people who had actually contracted the strain and needed help.
As such we wanted to work on reducing this stress by automating the testing procedure with the power of AI.
We believe that with an appropriate amount of data, deep learning algorithms could automatically analyze computed tomography (CT) of the chest and the clinical history, prioritize radiology worklists, and reduce the time for the diagnosis of COVID-19. Importantly, early diagnosis of COVID-19 is of critical clinical importance since the rate of morbidity is very high and the surge in the number of cases has been constantly increasing, and earlier treatment improves outcomes as well as alleviating the burden on the healthcare system.
What it does
Our Project has 3 stages that it operates in.
First and foremost is the data collection. The patient and the doctor can decide on the way to collect the required data. They can either use the very detailed written form, or the patient can talk about their habits and symptoms and our speech recognition algorithm will take care of the rest.
The app then uses OpenAI to compare the data against many recorded backgrounds and symptoms to give a prediction of the Covid test Result.
In case of a high probability of a positive result, the hospital technician can issue a CT scan of the patient's chest. Which can be fed into our convolutional neural network, to automatically decide whether the patient has covid or not.
Eliminating the need for a specialized doctor like a radiologist in most cases.
Patients and Technicians can obtain the results at any point in the process through a PDF.
How we built it
We first had to look for a particular problem to tackle in the current healthcare industry and given the scenery, Covid solutions made the most sense. Furthermore, Covid has affected so many aspects of everyone's life that it was hard to narrow down to one particular challenge to tackle. That's when the daily text of the UTD health check reminded us of how challenging the testing for Covid is since a significant percentage of the carriers can be asymptomatic.
That is when we decided to pursue the idea of CT scans as an option to identify Covid positive population. We started with a research of the background and symptom data that is necessary to make the correct decision since a lot of diseases can cause very similar symptoms. The next step was to translate this data into actual results, which with the help of open ai became easy.
The next step was to create technicians side of the project, with the ability to upload CT scans, and training the CNN to be able to detect the minute differences between the scans of a Covid patient and a Healthy person, fortunately, the internet had extensive data on the subject.
Challenges we ran into
One of the greatest challenges we had was the lack of knowledge about the healthcare industry.
As a team of two Computer Science students, we had very limited knowledge and understanding of the healthcare industry's response to the Covid virus. Thus finding a part of the problem that could be automated posed a bigger challenge than expected.
Another big challenge we faced was, how to obtain the required data while also maintaining the anonymity of the patients and following the HIPPA guidelines? We overcame this challenge by taking no record of the identifying traits of the person.
Accomplishments that we're proud of
We are proud of our easy-to-use interface, and the accessibility options with speech and tying inputs.
Our Neural Networks are also very robust and accurate.
What we learned
We learned the different testing techniques for the Covid virus, and also what symptoms distinguish Covid from any similar virus or disease.
What's next for COV-AI-D
We strongly believe that the new era of big data and artificial intelligence can play a pivotal role in overcoming many problems. Several strides have already been made to lend a helping hand in the diagnosis, identification, and risk-evaluation of a patient. However, these solutions as well as those in the coming future are heavily dependent on the availability of data. We aim to extend our reach to individuals who may have relevant information about any patient including demographics, history, experiences, symptoms, and medical documents. A comprehensive set of fields will enable us to gather fresh data and observe new patterns. We hope that our efforts will contribute towards improving existing technologies or perhaps sow the seeds of the next big idea.