Pulse excitation mode selection via AI Pipeline to Fully Automate the WUCT System

TL;DR

This paper presents an AI-driven method for optimizing ultrasound excitation pulse widths in CT systems, improving reconstruction quality without prior object information, using intelligent placement and deep learning evaluation.

Contribution

It introduces a novel AI pipeline that automates pulse mode selection and object placement, enhancing ultrasound CT reconstruction quality without needing prior data.

Findings

Deep learning model achieved 95.72% segmentation accuracy.

Optimized pulse width improves reconstruction quality.

AI pipeline automates parameter selection effectively.

Abstract

The parametric optimization for the ultrasound computed tomography system is introduced. It is hypothesized that the pulse characteristic directly affects the information present in the reconstructed profile. The ultrasound excitation modes based on pulse-width modifications are studied to estimate the effect on reconstruction quality. Studies show that the pulse width affects the response of the transducer and, thus, the reconstruction. The ultrasound scanning parameters, mainly pulse width, are assessed and optimally set by an Artificial Intelligence driven process, according to the object without the requirement of a-priori information. The optimization study uses a novel intelligent object placement procedure to ensure repeatability of the same region of interest, a key requirement to minimize the error. Further, Kanpur Theorem 1 is implemented to evaluate the quality of the acquired projection data and discard inferior quality data. Scanning results corresponding to homogeneous and heterogeneous phantoms are presented. The image processing step involves deep learning model evaluating the dice coefficient for estimating the reconstruction quality if prior information about the inner profile is known or a classical error estimate otherwise. The models segmentation accuracy is 95.72 percentage and intersection over union score is 0.8842 on the validation dataset. The article also provides valuable insights about the development and low-level control of the system.