A computer-aided diagnosis system of nuclear cataract.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Cataracts are the primary cause of global blindness, with nuclear cataract being the most prevalent form.
• Accurate grading of nuclear cataract severity is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To develop and validate an algorithm for the automatic diagnosis and grading of nuclear cataracts from slit lamp lens images.
• To assess the system's accuracy in detecting anatomical structures and predicting cataract severity compared to clinical standards.

Main Methods:

• A modified active shape model was used for automatic detection of anatomical structures within lens images.
• Local features were extracted based on clinical grading protocols using identified anatomical landmarks.
• Support vector machine regression was employed for predicting the nuclear cataract grade.

Main Results:

• The system achieved a 95% success rate in detecting anatomical structures, including the nucleus region, automatically.
• The average grading difference was 0.36 on a 5.0 scale, demonstrating high accuracy in severity prediction.
• Validation was performed on over 5000 clinical images with established ground truth.

Conclusions:

• The developed automatic diagnosis system for nuclear cataracts demonstrates high accuracy and reliability.
• This technology can enhance grading objectivity, reduce ophthalmologist workload, and support clinical and population-based studies.