In two newly published studies in Nature Biomedical Engineering, researchers trained a neural network to diagnose congenital cataracts with the same level of accuracy as individual ophthalmologists. Congenital cataracts is a rare disease that results in clouding of the eye’s lens and is responsible for about 10% of childhood vision loss worldwide. The AI, a novel method for diagnosing this disease, is the product of a multi-hospital effort spearheaded by Dr. Erping Long and Dr. Haotian Lin from Sun Yat-sen University in China.
A convoluted neural network is a computer algorithm that mimics our visual cortex, which allows the computer to learn to identify objects and specific visual motifs. Dr. Liu and his colleagues “trained” the convoluted neural network by exposing it to large sets of images, and telling the program which were patients with cataracts and which were controls in order for the AI to develop “rules” or cues for recognizing a patient with cataracts.
Producing an AI with specific image recognition capabilities requires hundreds of thousands of images pre-diagnosed for congenital cataracts and controls. All of this data is fed to the computer and the neural network undergoes a process called “deep learning” where it calculates specific parameters with which to identify what it is told to watch out for.
Dr. Liu tested the AI by having it diagnose previously unseen patients in a clinical setting, identify individuals positive for cataracts from a set of 50 cases, and identify those with cataracts from a phase-I multi-hospital clinical trial. The AI was able to diagnose the cataracts, determine their severity, and suggest treatment with an accuracy of about 90%.
Neural networks are being applied in multiple areas of medicine and have the potential to diagnose a variety of other diseases.
Studies in Nature Biomedical Engineering: Comparative analysis of image classification methods for automatic diagnosis of ophthalmic images…; An artificial intelligence platform for the multihospital collaborative management of congenital cataracts…