Fast and Accurate Detection of Hematoma Boundaries in Transcranial Ultrasound Brain Imaging Using Non-Convex Total Variation Regularization and Frequency Component Layer Separation

• Published in: 2023 IEEE International Ultrasonics Symposium (IUS) pp. 1 - 3
• Main Authors: Baradarani, Aryaz, Shapoori, Kiyanoosh, Malyarenko, Jeff Sadler Eugene, Gelovani, Juri G., Maev, Roman Gr
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.09.2023
• Subjects: Feature extraction; Image segmentation; Neuroimaging; Real-time systems; Three-dimensional displays; Transducers; Ultrasonic imaging
• ISSN: 1948-5727
• DOI: 10.1109/IUS51837.2023.10307342

The emergence of under-skull transcranial ultrasound (TUS) imaging holds promise for revolutionizing quantitative ultrasound diagnostics, providing a cost-effective alternative to conventional modalities such as CT scans and MRI. Medical ultrasound (US) imaging stands as a remarkable technological advancement, showcasing merits including portability, affordability, safety, fast imaging capabilities, and consistent diagnostic quality [1], [2], [5], [6]. TRUBI is a 3D transcranial ultrasound brain imaging system developed by Tessonics ® [7], [9], [10], [12]. It aims to overcome the limitations of traditional transcranial imaging, specifically the distortions caused by the human skull. Beyond the hardware and interface software aspects, the design and analysis of robust and consistent signal and image processing units for the system present a notable challenge. Within TRUBI, one of its processing units deals with the exploration of hematoma boundary detection, enhancing the segmentation execution. While the precision of the extracted boundaries remains pivotal, the system's ability to perform in near real-time demands due attention. In this context, we propose a new method to efficiently extract desired features with the aim of detecting hematoma boundaries during transcranial ultrasound brain imaging. Recent studies have demonstrated that non-convex total variation penalty can be defined in terms of the generalized Moreau envelope of a convex function, rendering the total cost function to be minimized in a convex form. Although the primary purpose of total variation is to estimate piecewise constant signals affected by noise, we employ this concept for the separation of signal components. The approach significantly mitigates the limitations observed in prior methods, enhancing accuracy while addressing computational complexity.