Title: Towards Practical Ultrasound AI Across Real-World Patient Diversity

Abstract:

In the case of high-tempo, traumatic scenarios on the battlefield, real-time ultrasound (US) imaging serves as an enabler for countless possible robotic interventions. Having the ability to automatically segment anatomical landmarks in the body, such as arteries, veins, ligaments, and veins, for percutaneous procedures remains to be a difficult task, even in “well-controlled” settings, such as hospitals, and therefore such procedures may experience complications. In this thesis, we focus on the task of semantic segmentation to address two key challenges when dealing with ultrasound image data: (1) extremely noisy images, which often leads to training data being expensive to collect, and (2) immense variations across ultrasound scanners, imaging settings, body types, and injury scenarios.

In our first work, we study how semantic segmentation neural networks can generalize across different populations of ultrasound images using transfer learning and propose a novel method for determining the optimal fine-tuning strategy to use to best achieve domain generalization. Our second work introduces novel temporal data augmentation strategies to increase the size of training data, specifically for dealing with the generalization setting of ultrasound scanning patterns. We evaluate our methods on multiple types of scanning patterns and notice improvements with our simple stochastic augmentation method.

The work following that focuses more on addressing the variations across body and injury types when ultrasound imaging them. We introduce a novel spatial non-uniform data augmentation method which is able to deform various sections of the ultrasound images to mimic long-tailed scenarios. Our final work introduces an initial prototype for a robotic system to automatically insert a needle into the femoral region of a patient. We integrate together various pieces of our semantic segmentation neural networks along with our studies on generalization for this work.

Committee:

John Galeotti (advisor)

Howie Choset (advisor)

Deva Ramanan

Chih-Kuan Yeh

Zoom Link: https://cmu.zoom.us/j/6078338716

Meeting ID: 607 833 8716