Analyzing and/or diagnosing an illness is no small task. Making the right call can mean all the difference to the ill and their family, while being wrong, can cause grave consequences. Because of the importance of accuracy in the healthcare industry, everyone from doctors, to hospitals, to insurance companies are relying more and more on the latest technological advances. Two areas in particular, Artificial Intelligence and Machine Learning, are helping doctors better predict illness and progression of recovery. In this blog, we’ll reference a Genetech and Roche development that is using AI to help predict when a patient with vision problems may go blind due to diabetes.
Where AI Predicts Vision Loss
Although doctors are currently able to estimate risks for patients with similar symptoms, they are unable to predict the sequence of vision loss in an individual patient. Genetech and Roche sought to use a model that could predict who is at a higher risk of losing sight within two years by using color photographs of the inside of the eye. Originally tested on patients with diabetic retinopathy, the model used deep convolutional neural networks (DCNN) to examine images and yield predictions. DCNNs work by assigning importance to a number of different objects in an image. The algorithm in this study looked for hemorrhages and microaneurysms, in an attempt to identify patients at high risk of going blind. Artificial intelligence was used to alert both patients and doctors of the risks at hand, while coming up with a strategy for monitoring and prevention of vision impairment.
Artificial Intelligence’s Accuracy
A recent study performed by the University Hospitals Birmingham NHS Foundation Trust and led by Professor Alastair Denniston compared AI performance with medical professionals. Examining both specificity and sensitivity, the results showed that AI and medical professionals are extremely similar when it comes to predictions and specificity. Analysis concluded that AI correctly diagnosed diseases 87% of the time while healthcare professionals were at 86%. Regarding the specificity of deep learning algorithms, AI was 93% accurate with professionals at 91% accuracy. While the stats paint a picture of the advancements of AI, more research must be completed. Current studies suffer from poor reporting, inconsistent terminology and lack of validation in real world outcomes such as timely treatment, hospital discharges and survival rates. It will be of upmost importance to continue these studies while mimicking regular clinical practices to gain true insights on the accuracy of AI in the healthcare industry.
Where We Stand Today
While there is continual pressure to develop new life-changing diagnostics and technology, the need for quality evidence of the benefits to patients and healthcare professionals is crucial. As AI and Machine Learning continue to mature and provide resources for healthcare professionals, it will take real-life, positive results to designate this technology as the final decision-maker in the process. AI is not going away anytime soon and we look forward to seeing the next advancements for use in the healthcare industry.
Key Information Systems