Purchase this article with an account.
A.C. Fisher, A. Chandna, I. Cunningham, D. Stone; An Expert System for the Differential Diagnosis of Vertical Deviation Strabismus . Invest. Ophthalmol. Vis. Sci. 2006;47(13):2464.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The differential diagnosis of vertical strabismus is routinely performed in clinic using the Prism Cover Test (PCT). The magnitude of the deviation, ideally in each of 10 cardinal positions of gaze, is recorded as the power of the neutralising prism. Frequently, not all positions are available, particularly in children. Prism powers are typically in the range +/– 20 prism dioptres (ΔD) at 0.5 ΔD resolution. The number of possible combinations and permutations is enormous. Consequently, the differential diagnosis into one of 8 types requires the interpretation of these data by an experienced clinical expert.
In this study, we enshrine this human knowledge into a machine–based (computer) artificial expert.
1. develop an artificially–intelligent (expert) method for the differential diagnosis and graphic representation of vertical strabismus; 2. characterise the potential redundancy in the 10 cardinal positions (with regard to differential diagnosis) to identify the minimum diagnostic set;
3. investigate how the requirements of the minimum diagnostic set can be decreased by semi–quantitative measures of eye movement.
A model for representing motor and sensory aspects of strabismus was designed to include essential aspects of assessment of strabismus into a single graphic. The starting point for the development of the model was the commonly accepted graphic representation of strabismus. An artificial dataset of 400 PCT examinations of 10 measurement fields was created. These were classified by the Clinical Experts into 8 classes of strabismus.
An artificial neural network (ANN) was constructed in MatLab to classify 100 randomly–selected records (Training set) into each of the 8 conditions. The remaining data were further sampled into 2 equal sets of 50 as Validation and Test sets.
The minimum diagnostic set for each diagnosis was determined iteratively in a separate Training–Validation set, with and without eye movement data.
A simple ANN based on a 3–layer perceptron with back–propagation was 100% accurate using the 10 cardinal measures in the PCT. Minimum diagnostic sets were determined and their enhancement by measures of eye movement quantified.
The expert system is 100% accurate in differential diagnosis. Further it reveals the minimum number of measures required for each diagnosis and value of eye movement measures.
This PDF is available to Subscribers Only