A large number of papers are appearing in the biomedical engineering literature that describe the use of machine learning techniques to develop classifiers for detection or diagnosis of disease. However, the usefulness of this approach in developing clinically validated diagnostic techniques so far has been limited and the methods are prone to overfitting and other problems which may not be immediately apparent to the investigators. This commentary is intended to help sensitize investigators as well as readers and reviewers of papers to some potential pitfalls in the development of classifiers, and suggests steps that researchers can take to help avoid these problems. Building classifiers should be viewed not simply as an add-on statistical analysis, but as part and parcel of the experimental process. Validation of classifiers for diagnostic applications should be considered as part of a much larger process of establishing the clinical validity of the diagnostic technique.

Optical coherence tomography provides sections of tissues in a noncontact and noninvasive manner. The device measures the time delay and intensity of the light scattered or reflected from biological tissues, which results in tomographic imaging of their internal structure. This is achieved by scanning tissues at a resolution ranging from 1 to 15 μm. OCT enables real-time in situ imaging of tissues without the need for biopsy, histological procedures, or the use of X-rays, so it can be used in many fields of medicine. Its properties are not only particularly used in ophthalmology, in the diagnosis of all layers of the retina, but also increasingly in cardiology, gastroenterology, pulmonology, oncology, and dermatology. The basic properties of OCT, that is, noninvasiveness and low wattage of the used light, have also been appreciated in analytical technology by conservators, who use it to identify the quality and age of paintings, ceramics, or glass. Recently, the OCT technique of visualization is being tested in different fields of dentistry, which is depicted in the article.

IntroductionThe method for measuring intraocular pressure using the Corvis tonometer provides a sequence of images of corneal deformation. Deformations of the cornea are recorded using the ultra-high-speed Scheimpflug camera. This paper presents a new and reproducible method of analysis of corneal deformation images that allows for automatic measurements of new features, namely new three parameters unavailable in the original software.Material and methodThe images subjected to processing had a resolution of 200 × 576 × 140 pixels. They were acquired from the Corvis tonometer and simulation. In total 14000 2D images were analysed. The image analysis method proposed by the author automatically detects the edge of the cornea and sclera fragments. For this purpose, new methods of image analysis and processing proposed by the author as well as those well-known, such as Canny filter, binarization, median filtering etc., have been used. The presented algorithms were implemented in Matlab (version 7.11.0.584 - R2010b) with Image Processing toolbox (version 7.1 -R2010b) using both known algorithms for image analysis and processing and those proposed by the author.ResultsOwing to the proposed algorithm it is possible to determine three parameters: (1) the degree of the corneal reaction relative to the static position; (2) the corneal length changes; (3) the ratio of amplitude changes to the corneal deformation length. The corneal reaction is smaller by about 30.40% compared to its static position. The change in the corneal length during deformation is very small, approximately 1% of its original length. Parameter (3) enables to determine the applanation points with a correlation of 92% compared to the conventional method for calculating corneal flattening areas. The proposed algorithm provides reproducible results fully automatically within a few seconds/per patient using Core i7 processor.ConclusionsUsing the proposed algorithm, it is possible to measure new, additional parameters of corneal deformation, which are not available in the original software. The presented analysis method provides three new parameters of the corneal reaction. Detailed clinical studies based on this method will be presented in subsequent papers.