A Brilliant Example of How Automated Deep Learning is Reshaping the Healthcare Analytics Industry
Overview
- A group of researchers have used an automated deep learning system for detecting damage in knee joints
- The model was trained using classification CNN and tested on 175 MRI scans
- The ROC metric showcased was more than 91% and specificity & sensitivity came out to be around 80%
Introduction
Healthcare analytics in a booming industry. Recent applications of artificial intelligence and machine learning in healthcare include Scanning Brain Anomalies faster than humans, Predicting Heart Diseases by Retina Scans, etc. Google even designed a system to predict the likelihood of a patient’s death with 95% accuracy!
Recently, a research project led by Fang Liu alongside the University of Wisconsin School of Medicine and Public Health, focused on using an automated deep learning based system to improve patient care. The researchers have created a model that can accurately identify knee joint cartilage and detect wear or injury with impressive accuracy.
The team has used segmentation and classification convolutional neural networks (CNNs) to train the deep learning system. To test this model, a dataset consisting of MRI images from 175 patients was used. These patients underwent fat-suppressed T2-weighted fast spin-echo MRI.
As a part of the test, two separate evaluations were performed and ROC (receiver operating curve) was used to measure performance. Below are the results of the two evaluations.
|
ROC |
|
|
Evaluation 1 |
0.917 |
|
Evaluation 2 |
0.914 |
As you can see, the deep learning system showed an overall high accuracy. It had a ROC score of 0.917 for evaluation one and ROC score of 0.914 for the second evaluation. Also, a sensitivity of 84.1% and a specificity of 85.2 % was achieved during the first evaluation, followed by 80.5 % sensitivity and 87.9 % specificity in the second evaluation.
|
Specificity |
Sensitivity |
|
|
Evaluation 1 |
85.2 |
84.1 |
|
Evaluation 2 |
87.9 |
80.5 |
As mentioned by the experienced radiologists, the main drawback of examining MRI for articular cartilages is a comparatively lower sensitivity. Looking at the high sensitivity provided by the system, it is expected to be the reason for bringing this into the healthcare industry and use it for practical scenarios. But before it is fully implemented in clinical practice, the model needs to be optimized.
Our take on this
The healthcare industry is always in need of more manpower and various researchers are working on providing AI and ML systems to help improve the current state of affairs. After having models detecting brain anomalies, early signs of diabetes and heart diseases, we now have a deep learning-based system that detects injury in knee joints (with a higher sensitivity than experts).
There is still tons of skepticism around AI but the hope remains that it will be used to assist clinical experts, rather than replace them.
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!
An avid reader and blogger who loves exploring the endless world of data science and artificial intelligence. Fascinated by the limitless applications of ML and AI; eager to learn and discover the depths of data science.
