Arti?cial Neural Networks Application in Intraocular Lens Power Calculation

Abstract

This article deals with intra-ocular lens (IOL) power calculations during the cataract surgery. At present, IOL power calculated by formulas is usually able to provide acceptable results for the majority of the patients. The problem appears when any of input parameters have the value which is not normal in population distribution. Then the patient post-operative refraction result can inconsiderable deviate from intended target. This work describes approach how to preoperatively indicate which samples of a patient could be problematic in accurate IOL calculations by classification of Artificial Neural Networks (ANN). Small and long eyes are used to test the ability of ANN to classify input samples which are taken from pre-operative measurements to several groups which represent probable post-operative result. In our experiment, ANN classifies samples into two groups. The first group is for data samples with a probable result in positive ranges of diopter and second group is for negative ranges. The accuracy of ANN, in this case, is 94.1 %.

Domain-specific loss design for unsupervised physical training: A new approach to modeling medical ML solutions

“…Olsen and Hirnschall et al tried to overcome these limitations by incorporating raytracing into their models [11,21]. Recently, an increasing number of formulas like the Hill-RBF or Kane also incorporate concepts of machine learning into their predictions [1,6,7,10,24,29]. Due to proprietary interests, most of the details regarding the used methods are not published, however, descriptions of e.g.…”