DPU - Symposium 2022

Medica l Image Ana lysis & Ar t i f icia l Intel l igence Symposium 2022 Nr. 16: Dr. Florian Schwarzhans, PhD, MSc: Real time eye tracking and blinking compensation for artefact free acquisition of polarization sensitive OCT volumes Curriculum Vitae: Florian Schwarzhans completed his PhD in Medical Physics at the Medical University of Vienna with a specialisation in Informatics with a focus on Programming using C, C++, MatLab and Python. He also has a background in Electronics specializing in Biomedical Engineering. His research interests include medical image processing with a special focus on automatic graphbased segmentation algorithms, deep learning methods for both image classification and segmentation, and the development and implementation of parallel algorithms for medical image processing and analysis using CUDA. His current research activities include automatic DICOM data conversion, as well as Radiomics feature extraction and feature robustness analysis using Python, by systematically and automatically introducing specific variations to the outline of the segmented lesions. Abstract: Real time eye-tracking and blinking compensation for artefact-free acquisition of polarisation-sensitive OCT volumes A technical solution for eliminating movement artefacts during acquisition of volumetric PSOCT scans was developed. The system used was a custom-built prototype PS-OCT operating at 860nm with an A-Scan rate of 70kHz and a field of view of 28°x21° (1024 A-Scans x 250 B-Scans) with an integrated line scanning laser ophthalmoscope for tracking operating at 790nm and a refresh rate of 60Hz. A real-time tracking algorithm with sub-pixel accuracy based on frequency space registration was implemented on the graphics card using the CUDA framework on C++. The calculated spatial displacement as well as an automatically calculated quality parameter is sent to the scanner to compensate for the detected motion. Several subjects were scanned with and without tracking, with the image centred at various retinal locations. Accuracy was analysed by scanning a single B-Scan position 250 times while the subject focused on a circularly moving fixation target, with and without tracking turned on.