Summary
A medical research organization (MRO) wanted to save their geriatric neurologists’ time and make diagnostic testing for Alzheimer’s disease more convenient and accessible to more people. Their ultimate goal was to help people identify Alzheimer’s in its early stages using artificial intelligence (AI). They enlisted Evalueserve for this project.
Evalueserve used computer vision to score Clock Drawing Tests (CDTs). CDTs involve patients being told to draw a clock with the hands pointing at a particular time. The results are used to assess and diagnose neurological and cognitive impairments – in this case, Alzheimer’s disease.
The Challenge
The nonprofit needed help classifying and predicting the progression of Alzheimer’s. They wanted the scoring of CDTs to be automated. Traditionally, neurologists give CDTs a score between one and five, with five being the most severe. These scores are based on the accuracy of the time displayed and the overall state of the drawing. These tests diagnose Alzheimer’s and other neurological disorders and can be a tool for assessing the condition’s severity.
Our Solution
The objective was to automate the CDT scoring process by introducing AI – specifically, computer vision and deep learning – to the process. In addition to adding speed and scale to the process, computer vision can be more consistent than the human eye and less subjective than a doctor’s judgment. Automating the scoring process not only saves time for geriatric neurologists but also improves direct medical research.
To automate CDT scoring, Evalueserve took the following steps:
- Data Collection and Processing – Our experts collected images and their corresponding CDT scores and mapped them. The sample data was then extracted and annotated for use in model development.
- Feature Extraction – Our experts used identification and creation features to simulate human decision-making based on domain understanding. They developed a robust object detection pipeline (number recognition, clock hand recognition, clock circle model, number, clock hand position, etc.) to detect individual features and their corresponding locations using various AI algorithms (ResNet, OpenCV, Tensorflow Object Detection).
- Classification – Based on features, the images were then classified using a classification engine.
- Product Development – The AI engine was packaged into a Docker environment with a back-end Django engine for creating the web-based application. The AI engine could be called through an API engine.
With the AI in place to score the CDTs, the process is much the same as if a patient were to see a neurologist in person. From their own homes, patients access an online platform, follow the prompt to draw their clock, submit a photo of the clock drawing, and then receive the AI-powered score. As with in-person testing with a neurologist, they receive scores ranging from one to five. Based on their results, the patients can be diagnosed with Alzheimer’s or other conditions and informed of their condition’s current severity.
Business Impact
Our computer vision-powered CDT scoring solution helped the MRO benefit more people with its work. The CDT scoring solution allowed patients to be seen from the comfort of their own homes, which democratized the assessments, making it so the MRO could provide CDTs to more people at a lower cost. Their accessibility removed barriers in the way of people receiving Alzheimer’s diagnoses, hopefully helping more people catch the disease in its early stages. When caught early, there’s a possibility of slowing the progression of Alzheimer’s in different ways, bringing relief to both patients and their loved ones.
Our solution also gave patients more dignity in a sensitive situation. Finally, powering our solution with AI made its diagnoses more consistent, as it removed doctors’ subjectivity and varying experience and expertise levels from the equation.