Ultrasound based machine learning classification to match imaging properties of real tissues to artificial structures

Advancements in sensors, computing, and surgical data science have improved medical treatments, particularly minimally invasive procedures (MIPs), offering benefits like faster healing and reduced patient discomfort. However, challenges persist in terms of accuracy, safety, and awareness during MIPs, including difficulties in locating and tracking minimally invasive devices within the body. The research aims to address these challenges by developing sensor-based guidance methods and transformation algorithms, together with machine learning, achieving a mean F1 score of 0.8 in classifying ultrasound images of tested materials. This represents a step toward real-time tissue differentiation, ultimately offering safer and more accurate MIPs.Clinical relevance: This paper enhances MIPs by creating realistic interventional phantoms for training and device testing, reducing errors improving patient safety, and integrating machine learning classifications for real-time guidance.