Gal Alon- DETECTION OF CORPUS CALLOSUM MALFORMATIONS VIA SPATIO-TEMPORAL LATERAL FACILITATION MODEL

You are invited to attend a lecture on Tuesday, January 23^rd, 2024

15:00

Seminar room 011

DETECTION OF CORPUS CALLOSUM MALFORMATIONS VIA SPATIO-TEMPORAL LATERAL FACILITATION MODEL

By:

Gal Alon

M.Sc. student under the supervision of Dr. Hedva Spitzer

Abstract

Ultrasound B-mode examination is a non-invasive and safe way to perform prenatal examination. However, the ability to detect prenatal malformation by medical personnel sometimes is hampered due to the large multitude of artifacts and noise in the resulting video-image. Brain malformations can lead have significant developmental issues and early detection of such malformations can lead to a clinical decision on termination of pregnancy. In order to improve the diagnostic confidence, considerable effort is put towards improving the quality of the resulting image including speckle noise reduction, ultrasound image enhancement, and segmentation. In this work, we focus on a specific medical problem – the detection of malformation in a certain part of the brain, i.e the Corpus Callosum.  The propose of the algorithm that implements an existing line completion mechanism of the visual system, is to further elaborate  it to line completion across several frames to allow detection of moving element across time. The algorithm results were evaluated in a survey across 25 different ultrasound cases and 10 medical personnel. While the algorithm results show small improvements in high quality ultrasound video images, there was no significant improvement over the control group. We have shown a correlation between the algorithm's improvement and the ultrasound's noise reducing parameters (CRI, SRI). This correlation suggests that the underwhelming performance stems from highly structured "noise", which breaks the algorithm's assumptions that the only structured part of the video-image is the desired signal. It is also possible that the contribution of the temporal dimension to line-completion may be negligible in comparison to the contribution of the line completion algorithm in the image frame itself. This suggestion might further explain the resulting performance. Nevertheless, the algorithm presents a novel approach to the generalization of visual mechanisms to spatio-temporal domains and can be improved in several ways to hopefully yield satisfactory results.