Dignosys: The Future of Diagnosis

Original article was published by Aditya Mittal on Artificial Intelligence on Medium


Dignosys: The Future of Diagnosis

As a child, the thing that I always wanted to grow up to be was a doctor; I used to dress up as a doctor on Halloween and play Operation in hopes of experiencing the same thrill that doctors do when they operate. However, something that I had not fully comprehended at that time was the fact that the patients that doctors worked on were real. As I grew older, I realized the stress that doctors had to perform at the optimal level during operations, and I knew that they were bound to make some mistakes as human error can never be taken out of the equation.

Common reasons for misdiagnosis, including human error. (Image Source)

I decided to research more into how human error plays a role in hospitals. What I had found appalled me. 12 million misdiagnoses per year. Of that, 1 in 3 are fatal. Although doctors do commonly make the best decisions, each of these misdiagnoses affects an entire family, simply because the doctor is unable to make a definite assumption from the data provided.

“All of the tools are there to make a decision, but sometimes, using the results from all of those different [tests] … can be difficult for diagnosis” — Samvid Dwidevi, DO, Anesthesiologist, Henry Ford Hospital

The problem is: how can we provide a more accurate diagnosis from the data provided without relying on doctors? Enter Dignosys, an artificial intelligence tool that will allow any individual to receive accurate and constant medical diagnosis without the need for doctors.

Dignosys and How It Works

Dignosys provides a secure platform for analyzing patient data; unlike traditional hospitals where patients have to go in-person to see their doctors, Dignosys allows patients to communicate with doctors in a secure manner and receive the risk of contracting cardiovascular or respiratory disease from the AI algorithm that is built into the app.

Blockchain

The log-in page for doctors and patients through the website.

The system uses a decentralized system to safely transact data from the doctor to the patient. Our application provides a secure log-in for both patients and doctors, where doctors will be able to see all of their patients and their medical history, and patients will be able to see their own data. From this app, patient-level data that is added by the doctor will be subsequently stored on the blockchain but will only be available for the doctor and patient to see.

The user interface for a doctor checking his patients from the dashboard.

This system provides an easily-accessible way for doctors to manage their patients and a distanced version for patients to access their medical data. From the data, patients and doctors will be allowed to run the data through the AI model to determine the risk for cardiovascular or respiratory disease.

X-Ray Data

In hospitals, data comes in many different forms, ranging from tables to MRI scans to x-rays. Our app can currently analyze from x-ray and tabular data, but we are working towards integration of all medical data.

If the doctor chooses to diagnose from an x-ray image, the data will be run through a convolutional neural network that will analyze the data. Using this will provide ease to the radiology department, as medical image-detection training requires years of practice and is very difficult to master. A machine, on the other hand, can recognize the patterns relevant to the disease and alert the doctor or patient accordingly.

“There is always going to be human error, and humans interpret what the machine is telling us.” — Dr. Sylvia Delius, Radiology, Kaiser Permanente

The neural network will provide the risk of the patient having a variety of respiratory diseases (such as pneumonia and COVID-19), allowing patients to narrow down what exactly is the disease and how should they treat it. By doing so, patients and doctors can go forth with certain treatments depending on the diagnosis provided by the machine.

MRBase

The dashboard to input patient data to run through the model.

Our app integrates MRBase to determine the most significant risk factors that will help diagnose cardiovascular disease. MRBase is a repository of genetic-wide association studies that maps genetic variants to traits in the body; it uses two-sample Mendelian Randomization and a random-forest machine learning algorithm to determine how significant a specific risk factor is to a disease.

Before sending the data to the AI model, our program preprocesses the data by determine which columns are most relevant to cardiovascular disease. This is done so that irrelevant data is not passed into the model, thus skewing the final result. The relevant columns are then chosen to be passed into the ensemble learning model.

Ensemble Learning

The preprocessed data is then sent into an ensemble learning model that will determine the risk of cardiovascular disease. In this application, five algorithms (random forest, logistic regression, decision trees, multinomial naive bayes, and support vector machines) comprise the ensemble model. The model was trained using Cleveland Heart data from UC Irvine, and runs on thirteen features.

Similar to the convolutional neural network, this app will provide the risk of cardiovascular disease from various models to form one hypothesis, and it will be able to detect certain subsets of cardiovascular disease if necessary. This will greatly ease stress on cardiologists and increase the accuracy of diagnosis, paired with their input.

“It’s not always easy [to diagnose cardiovascular disease]. It takes a long time and a lot of dedicated training to recognize patterns and determine their risk of having the disease.” — Samvid Dwidevi, DO, Anesthesiologist, Henry Ford

Flask

To build the framework for the user interface, Flask, a Python micro web framework, was used to integrate the various components of the application. It was utilized to connect the AI model to the target result of the prediction that will be displayed on the application.

Additionally, using Flask, our application provides an easy-to-read user interface for both patients and doctors that will allow users to interact with data in an accessible manner. This will provide more ease to the patient than the conventional method of retrieving data from a hospital.

Dignosys, Simplified

  • Secure: Using decentralized computing systems to safeguard patient-level image and tabular data and maintain doctor-patient confidentiality.
  • Analyze: Using machine learning algorithms to sift through patient data and determine the risk for both cardiovascular and respiratory diseases.
  • Diagnose: Using an easy-to-read user interface that will give patients and doctors a more substantial need of a patient’s medical needs.

Conclusion

Using Dignosys allows hospitals to relieve stress from doctors and provide more accurate diagnosis. By implementing artificial intelligence and secure blockchain system, the application uses cutting-edge technology to provide the best experience and results to both the doctors and the patients.

For more information about our product, check out our website.