Automated user assessments of ultrasound transducer performance

Abstract

This work concerns the use of automated image analysis techniques that could be applied to help users assess the condition of clinical ultrasound transducers. A stakeholder questionnaire was completed to identify current user QC practice and identified that users often find existing equipment performance tests time consuming, and challenging to conduct. A literature review found that performing reverberation image assessments to check for element dropout or a non-uniform response across the probe array and changes in sensitivity over time are effective tests for users to perform. To address issues identified in the questionnaire, two automated image analysis techniques were developed and assessed; one performing analysis on patient images and the other on reverberation images. Python code was created to crop curved and linear B-mode patient images so that analysis techniques could be applied and evaluated. The code successfully cropped 567, 381, 21 and 5 GE ML6-15, L2-9, C1-6 and IC5-9 patient images and the study was the first to successfully crop and linearise curved patient images. Merging patient images allowed a known defect on a GE ML6-15 probe to be identified but this was not a successful assessment method on other probes as suspected defects were not detected in the merged pixel profile. Image analysis on patient images showed that the mean, coefficient of variation and kurtosis of pixels within patient images overall weren’t sensitive enough to detect known probe defects; T-tests on these parameters before and after a defect developed in the ML6-15 probe resulted in a p value of 0.51, 0.14 and 0.35 respectively. In contrast, the standard deviation did show a significant difference with a p value of 0.001. T-tests on the Benford law integer distribution for each image before and after the presence of dropout resulted in a p value >0.05 for integers 1-9. Benford’s distribution therefore showed no significant difference before and after the presence of dropout and isn’t a sensitive test. Python code was created to crop both linear and curved reverberation images and to perform uniformity and sensitivity analysis. The coefficient of variation (Ucov), skew (Uskew), a measurement of the lowest signal to measure signal loss (Ulow) and the reverberation depth (Sdepth) were successfully used to analyse the reverberation image. Normal ranges of were found to be between 1.2% and 5.4% and -0.1 to -3.1 for Ucov and Uskew respectively. Generally, on probes with acceptable uniformity, Ulow was found to be below 28.6% at the edges and up to 7.9% for the central regions. This was the first study to investigate the influence of temperature on the resultant reverberation image. Temperature was shown to affect the appearance of the reverberation image. An increase in the probe temperature resulted in peaks being shifted to deeper depths in the image. It also caused peaks to merge in some positions. This did not significantly affect the automated analysis of reverberation images but could impact sensitivity measurements when the probe temperatures for the test is significantly different to that when baseline images were captured.

Date of Award	31 Dec 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Karen Kirkby (Supervisor)