Improving the robustness of the Sequentially Optimized Reconstruction Strategy (SORS) for visual field testing

Area of Science:

• Ophthalmology
• Computational Vision
• Medical Imaging

Background:

• Perimetry measures light sensitivity across the visual field.
• Linear regression models can reconstruct visual fields from subset measurements.
• Improving reconstruction robustness is crucial for efficient visual field testing.

Purpose of the Study:

• To introduce a novel method for visual field reconstruction using dimensionality reduction.
• To evaluate the performance of the transformed-target principal component regression (TTPCR) method.
• To assess the impact of dimensionality reduction on reconstruction accuracy and efficiency.

Main Methods:

• Implemented principal component analysis (PCA) for dimensionality reduction of perimetry data.
• Developed the transformed-target principal component regression (TTPCR) algorithm.
• Trained and tested TTPCR against traditional linear regression on varying dataset sizes.

Main Results:

• TTPCR achieved comparable results to linear regression on large datasets.
• On small datasets, TTPCR required 22% fewer trials to reach similar error levels.
• No underfitting was observed with parameter reduction in TTPCR.

Conclusions:

• Dimensionality reduction techniques, like PCA, enhance the robustness of visual field reconstruction algorithms.
• TTPCR offers improved efficiency, particularly in scenarios with limited training data.
• This approach holds potential for more efficient and reliable perimetry.