The goal of this clinical trial is to test the performance of neuronal networks trained on ultrasonic raw Data (=radiofrequency data) for the assessment of liver diseases in patients undergoing a clinical ultrasound examination. The general feasibility is currently evaluated in a retrospective cohort.
The main questions the study aims to answer are:
* Can a neuronal network trained on RF Data perform equally good as elastography in the assessment of diffuse liver diseases? * Can a neuronal network trained on RF Data perform better than a neuronal network trained on b-mode images in the assessment of diffuse liver diseases? * Can a neuronal network trained on RF Data distinguish focal pathologies in the liver from healthy tissue?
To answer these questions participants with a clinically indicated fibroscan will undergo:
* a clinical elastography in Case ob suspected diffuse liver disease * a reliable ground truth (if normal ultrasound is not sufficient e.g. contrast enhanced ultrasound, biopsy, MRI or CT) in case of focal liver diseases, depending on the standard routine of the participating center * a clinical ultrasound examination during which b-mode images and the corresponding RF-Data sets are captured
scheduled for an ultrasound investigation by an independent physician
signed declaration of consent
smaller interventions in the same liver during the last 2 Week (for example liver biopsy)
contrast enhanced ultrasound less than a day ago
major intervention at the liver (for example partial resection)
Parallel
Collection of elastography data
Devicepatients who are scheduled for an elastography for clinical reasons usually receive an ultrasound scan in which the b-mode images of the liver tissue are collected. In this study additional radiofrequency data is collected through a software access.
Collection of ultrasonic raw data
DevicePatients who are transferred to the ultrasound departement due to suspicious focal lesions receive an ultrasonic investigation including the acquisition of raw data and extracting a definitive diagnose from the following clinical routine investigation, depending on the standards of the participating center
Performance analysis of the trained model
Analysis of the concordance of a Deep Learning-based analysis of RF data with established clinical measures. In case of diffuse disease the stiffness of the tissue and in case of the focal lesions the underlying disease as diagnosed by the local physicians are the measures. Performance is evaluated by the area under the receiver operating characteristic curve and a correlation coefficient.
Click on the lead sponsor to view all of their trials.
Technische Universität Dresden
Lead sponsor
University Hospital Dresden
Collaborator
University of Leipzig
Collaborator
