Deep Learning an Artificial Neural Network

Over 400 million people have diabetes today. Due to several hereditary and lifestyle factors, many individuals tend to suffer from diabetes from a very young age. The number of patients has doubled in the past 30 years, especially in Asia. One-third of this population is adversely affected by Diabetic Retinopathy- a medical condition in which the retina is damaged due to diabetes mellitus.

Diabetic retinopathy usually does not show any symptoms, until the retinal tissue begins to swell and could cause visual impairment if not diagnosed at an early stage. This rising concern has made the world’s eye turn towards finding a cure for this condition.

Early diagnosis of diabetic retinopathy is particularly difficult as it involves the detection of microaneurysms, retinal hemorrhages and ruptured blood vessels by fundoscopy. Recently, the terms ‘artificial intelligence’ and ‘machine learning’ are becoming very popular. When the machine performs certain tasks by itself which usually require human intelligence- Artificial intelligence. Machine learning is the ability of the machine to teach or improve itself, without the requirement of prior programming. Google has recently developed an AI Module that will help in the early diagnosis of diabetic retinopathy, along with several initiatives in India. This algorithm is an example of Deep learning- an artificial neural network designed based on the human brain.

Google team had to first obtain many images captured by the fundus camera of different cases of this condition. The Eye Patient Archive Communication System, in the US and Arvind Eye Hospital, Sankara Nethralaya ,and Narayana Nethralaya- hospitals in India were used to retrieve this data. Many number of ophthalmologists graded these images according to the International Clinical Diabetic Retinopathy scale of none, mild, moderate, severe or proliferate. The machine was initially made to guess what it might be and then compared to the diagnosis made by the ophthalmologists. It does not include pre-written programs to detect retinal hemorrhage or cotton wool spots. The severity of this condition is detected as the machine reads every pixel of the image. Certain characteristics that differentiate the healthy and diabetic eye are mapped into a data set. A statistical approach is then adopted to classify the data sets based on the previously made inferences.

Although this algorithm still requires FDA approval, it’s future is very bright and can help prevent visual impairment by reducing the volume of people suffering from diabetic retinopathy. It may also help reduce the overall cost of eye disease diagnosis. It may also find its application in the detection of ocular melanoma in the future.

 

Cite this page