Design and Implementation of Anomaly Detection through Deep Residual Network

Abstract

The detection of diagnostically relevant markers in imaging data is critical in medicine and treatment guidance. This makes large scale annotation necessary which is often not feasible.  In this paper, we are designing a robust detection model for anomaly localization in Retinal OCT images. OCT refers to Optical Coherence Tomography which produces cross-sectional images of the retina using the low coherence interferometry[4],[6]. The existing methods consist and consider normal dataset group only and also detects the deviation as an anomaly from the normal data. To overcome these issues, we propose an in-depth analysis to detect an anomaly. OCT consists of Super Luminescent Diode (SLD) as a light source wherein an infrared light is used[8]. The diode splits light into two parts, the first sent to sample and another to the reference arm with a known wavelength and an interferometer is used to study the back waves from the reference arm. The proposed method promotes the industrial anomaly detection as a part of its application. This technique of detecting an anomalous behavior helps in gaining useful insights in various industrial application. OCT artifacts could be related to patient, operator or software, OCT is used mainly for diagnosis, and for checking the functioning of optic nerve and retinal disorder[10],[14]s. Diseases like Choroidal Neovascularization, causing painless loss of vision, Diabetic Macular Edema, causing wavy vision along with washed out or faded color and Drusen, causing Metamorphopsia which are prevalent among the age group of 45-60 can be predicted[2].