Background: Efficacy and high availability of surgery techniques for refractive defect\ncorrection increase the number of patients who undergo to this type of surgery.\nRegardless of that, with increasing age, more and more patients must undergo cataract\nsurgery. Accurate evaluation of corneal power is an extremely important element\naffecting the precision of intraocular lens (IOL) power calculation and errors in this\nprocedure could affect quality of life of patients and satisfaction with the service provided.\nThe available device able to measure corneal power have been tested to be not\nreliable after myopic refractive surgery.\nMethods: Artificial neural networks with error backpropagation and one hidden\nlayer were proposed for corneal power prediction. The article analysed the features\nacquired from the Pentacam HR tomograph, which was necessary to measure the\ncorneal power. Additionally, several billion iterations of artificial neural networks were\nconducted for several hundred simulations of different network configurations and\ndifferent features derived from the Pentacam HR. The analysis was performed on a PC\nwith Intel�® Xeon�® X5680 3.33 GHz CPU in Matlab�® Version 7.11.0.584 (R2010b) with\nSignal Processing Toolbox Version 7.1 (R2010b), Neural Network Toolbox 7.0 (R2010b)\nand Statistics Toolbox (R2010b).\nResults and conclusions: A total corneal power prediction error was obtained for\n172 patients (113 patients forming the training set and 59 patients in the test set) with\nan average age of 32 �± 9.4 years, including 67% of men. The error was at an average\nlevel of 0.16 �± 0.14 diopters and its maximum value did not exceed 0.75 dioptres. The\nPentacam parameters (measurement results) providing the above result are tangential\nanterial/posterior. The corneal net power and equivalent k-reading power. The analysis\ntime for a single patient (a single eye) did not exceed 0.1 s, whereas the time of network\ntraining was about 3 s for 1000 iterations (the number of neurons in the hidden\nlayer was 400).
Loading....