Fig. 1: Slit-lamp images of a normal eye lens (left) and a lens with nucelar cataract (right)
As people age, water-soluble proteins within the lens of the eye may lose transparency and become cloudy. This condition, known as cataract, is the leading cause of blindness worldwide. Recent studies in the United States have estimated that approximately 17% of adults over 40 and more than half of adults over 75 have cataracts or have had cataract surgery.
There are several types of cataract related to aging. Nuclear cataract, which begins at the center of the lens and progresses towards the surface, is the most common type. To diagnose nuclear cataracts, doctors take images of the eye lens using a slit lamp machine, and then assess the lens opacity and color using a grading system such as the Wisconsin cataract grading system. However, the entire clinical assessment process has so far been manual and subjective — the results of which can vary substantially among practitioners.
Huiqi Li at the A*STAR Institute for Infocomm Research in collaboration with the Singapore Eye Research Institute have now developed a computer-aided diagnosis system for grading nuclear cataract. The system, known as automated grading of nuclear cataract (AGNC), has several unique features. First, the AGNC system automatically recognizes the lens structure, in particular, the nuclear region of the lens. It uses 38 dots, or landmark points, to outline the contour of the lens structure. It also has a user intervention function for images that the AGNC system cannot handle due to inaccurate focus, small pupils or a dropping upper lid.
The AGNC system extracts the intensity, color, texture and other features within the nucleus region that are useful for grading. It even measures the intensity of the sulcus, one of the most important factors that reflect the severity of nuclear cataract.
The researchers have validated the performance of AGNC in accurately grading the lens opacity and color against more than 5,000 slit lamp images obtained from clinics. The system automatically processes 95% of the images without any user intervention. More importantly, the grading difference between the AGNC system and human graders for 97.5% of the slit lamp images was less than one grade (on the Wisconsin cataract grading system), which is very acceptable for clinical diagnosis purposes.
The quality assurance demonstrates that the AGNC system is capable of diagnosing nuclear cataract automatically and accurately. “The AGNC system provides an automated, precise and quantitative assessment of nuclear cataract,” says Li. The researchers are currently testing the performance of the AGNC system in clinical environments.
The A*STAR-affiliated researchers contributing to this research are from the Institute for Infocomm Research.
