This publication was both a Master’s thesis as well as a PhD research prospectus, with the subject area chosen specifically to best guide the efforts at SonoSim, Inc. in enhancing the SonoSim® Ultrasound Training Solution and the overall SonoSim experience.

Surveyed the field of semantic biomedical image segmentation, identifying key trends in recent deep learning research and forecasting the field’s future trajectory. Contrasted modern ensemble models and multi-scale, multi-stream volumetric segmentation network architectures, elucidating the contributions of each topology to the current state-of-the-art.

Using ultrasound as a motivating example, articulated the future applications enabled by automated image analysis algorithms and the corresponding idiosyncratic challenges faced in attempts to provide these capabilities. Demonstrated the segmentation-dependence of these methods and the resultant impact of expert-level segmentation models across many contexts, from the classroom to the clinic. Underscored the manifold ways in which simulation-based ultrasound education and training in conjunction with AI-driven solutions can address the still largely unfilled need for qualified ultrasonographers across many medical specialties and underutilized use-cases.

Teofilo E. Zosa. "Catalyzing Clinical Diagnostic Pipelines Through Volumetric Medical Image Segmentation Using Deep Neural Networks: Past, Present, & Future." UCSD Computer Science and Engineering PhD Research Mastery Exam, June 7, 2019.*

* Survey paper ranked "at the level of a good paper or presentation at a top conference in the area of the exam", the rarest and highest possible examination rating.