Skip to main content

Contrast-enhanced uLtrasound for livEr-disease eValuation: development and validation of a novel E-Health-software for Risk-stratification

Final Report Summary - CLEVER (Contrast-enhanced uLtrasound for livEr-disease eValuation: development and validation of a novel E-Health-software for Risk-stratification)

The main objective of the research proposed for this TOK action is to develop a completely novel automatic tool for the analysis of the liver vascular network by modeling DCE-US images, able to analyze functional properties and create vascular maps. This objective must be broken down into four specific objectives:
1.) To develop automated DCE-US tools for non invasive assessment of vascular networks of the liver. 2.) To develop advanced graph analysis tools for improved modeling of the patient. 3.) To develop imaging biomarkers&predictive models that enable direct understanding of patient’s risk. 4.) To integrate the above solutions in a final prototype
THE PROBLEM: Liver vascular network is characterized by a highly organized structure. This is progressively deranged due to fibrosis and hepatocyte damage in patients with chronic liver diseases. Non-invasive methods able to quantify objectively&automatically the degree of liver vascular derangement, are needed.
OUR SOLUTION: Temporal analysis of Dynamic Contrast Enhanced Ultrasound (DCE-US) based on non-invasive imaging enables to obtain a personalised graph model of the vascular network of the liver. We hypothesised that this vascular network model contains relevant information to stratify risk of individuals and support clinical decisions in a objective manner.
FUTURE: We foresee the expansion of this technology to other fields of biomedicine, so we are now developing a new solution applied to fertility which has provided successful results on longitudinal studies for 126 individuals.

A description of the work performed since the beginning of the project
The two academic institutions worked together with Ymaging to develop a large database of well-characterised patients with liver disease in form of electronic Case Report Form (eCRF) receiving liver DCE-US acquisitions according to the CLEVER protocol. Ymaging provided the technology enabling the secure online transmission of the videos of DCE-US acquisitions to the central server from both institutions. Clinical information was securely stored after anonymisation and systematically backed-up, and linked to the video sequences acquired.
Videos were initially analysed according to the algorithms provided by the background methods (Amat-Roldan et al. Radiology 2015). Additionally Ymaging played a critical role in several WPs by carrying work on the integration of CLEVER software as web service available to members of the consortium in a secure manner;..

Development and refinement of the CLEVER software
A total of 387 patients with approximately 900 video acquisitions of the liver were prospectively enrolled. All Milestones of the project have been fully achieved. Main aspects that were improved along the project are: (1) data conditioning for types of ultrasound scanners (Acuson, Supersonics, Esaote...) to adapt different DICOM formatting as well as basic image processing; (2) motion detection and correction was successfully improved and tested on 2 scanners so far (Acuson and Esaote); (3) time series analysis which is the base of graph mapping was extended by different means from parametric fitting of temporal curves to link estimation based on different temporal characteristics; (4) advanced feature extraction and machine learning techniques were used to obtain families of selected features that were informative of derangement of liver microcirculation; (5) additional machine learning means were further implemented to build reliable regression models in different HVPG ranges according to the available sample size available for training; (6) additional covariable analysis combined to machine learning enable to build unified model to provide a reliable and single HVPG value estimated from CEUS video analysis, platelet count and spleen diameter; (7) successful blind testing enabled to obtain high correlation values, particularly with Acuson.
The CLEVER online software now estimates the HVPG by offline graph model analysis of the vascular connectomes on the microbubbles reperfusion images of Acuson and Esaote. The pipeline then provides the CLEVER HVPG estimation in mmHg, together with a grading of reliability of this estimation (“poorly reliable”, “reliable” or “highly reliable”). 69 patients with single DCE-US acquisitions with Acuson were used as blinded validation set. A similar approach was used to implement the pipeline for the videos acquired with Esaote at UNIBO.

Socio-economic impact
The problem
Liver cirrhosis is the severe life threatening condition representing the final stage of chronic liver disease regardless of the cause (alcohol, hepatitis B and C virus, non-alcoholic steatohepatitis, hemocromatosis,...). Chronic liver disease occurs in all Countries worldwide and affects approximately 6% of European Union (EU) citizens, with an estimated mortality of 14.3 per 100.000/ in EU in 2004 (5th most common cause of death in the EU). Early recognition of the severity of the disease and precise prognostic information to the patient helps delivering appropriate management strategies, including liver transplantation. .

State-of-the-art
Current techniques for liver evaluation and stratification include Hepatic venous pressure gradient (HVPG), wihch is the reference method to estimate portal hypertension and is obtained by hepatic vein catheterisation. This means that it is an expensive minimally invasive technique (complications <1%, cost in Spain 1600€). Moreover, the availability of facilitities where HVPG is carried out with adequate skills is rather limited and tends to concentrate only in referral centers. HVPG is however a strong independent prognostic indicator in compensated and decompensated cirrhosis. It is also a key variable to decide upon surgical operability in case of liver cancer in cirrhosis. Current alternatives are lacking. The accuracy of laboratory tests for diagnosing clinically significant portal hypertension does not exceed 60-70%. Shear Wave Elastography is a well validated technique able to assess tissue stiffness and has been tested in recent years as a novel way of obtaining numerical, objective and operator-independent noninvasive surrogate data of HVPG. However, the correlation of TE with the HVPG is unsatisfactory (diagnostic accuracy around 80%).

Our solution and their benefits
The outputs of the CLEVER allow a personalised and simple to perform prognostic stratification of patients with cirrhosis integrating or replacing HVPG. The goal was achieved with collaboraton of participants partners, who are opinion leaders.

How does CLEVER solution works?
An ultrasound contrast agent (microbubbles) is injected in a vein of the patient spreading in the bloodstream. Microbubbles can be imaged in real time with high accuracy by most mid or high end conventional ultrasonographic equipments provided with contrast-enhanced ultrasound specific software. By imaging the liver by ultrasound the microbubbles can be selectively visualized and disrupted by specifically devoted ultrasound emissions when needed. After disruption, additional microbubbles naturally replenish the liver via the bloodstream depicting the hepatic vascular network. This cycle is recorded and analyzed with the entire procedure lasting <15 minutes. CLEVER researchers developed a software that automatically assessesses the liver microcirculatory network and provides a non-invasive estimation of portal pressure (HVPG). The main advantages are that this system could reduced costs for portal pressure measurement and extend the possibility to estimate portal pressure to a very large number of patients worldwide also in locations where HVPG measurement is not in place.
More info at: https://vimeo.com/122526192

Results of the CLEVER software in the included population
Among the Acuson sequences analysed up to the present moment, the CLEVER software has been able to provide portal pressure estimations from DCE-US videos in 85/152 patients (56%). Applicability increased when > 2 videoclips were acquired: 51% with one acquisition,56% with 2,81% with 3, and 100% with 4.
In the training set 28/50 patients with CLEVER results had reliable or highly reliable CLEVER values and showed excellent correlation with HVPG (range of HVPG 2.5 to 26 mmHg): R= 0.914 p<0.0001. In the blinded validation set 21/35 patients had reliable or highly reliable results confirming a very high correlation r=0.834 p<0.0001 (Figure 1). These results show that the CLEVER software is unique in allowing an accurate non-invasive estimation of portal pressure in patients with chronic liver disease. Exploiting larger sample size available now for training, we expect correcting and increasing correlation in a relevant manner. Increasing such correlation will also have a direct impact on video acceptance while keeping enough clinical precision. Similar results were obtained in patients included at UNIBO. The technical applicability was approximately 50% of videos in the training set regardless of whether taken from the same or different patients (41/90 videos, corresponding to almost 60% of patients, similar to Acuson) forecasting a quite high applicability when multiple video acquisitions will be taken from the same patient in the future. The correlation of the CLEVER estimate of portal pressure with HVPG was 0.585 (p<0.001) and increased to 0.701 (p<0.002) in the blind validation set.

These results are object of two separate abstracts, respectively one submitted to the International Liver Congress of the European Association for the Study of the Liver (EASL) to be held in Paris in April 2018, the second to the Annual Meeting of the Italian Association for the Study of the Liver (AISF) to be held in Rome, in 2018.


www.cleverpoject.eu

Relevant contacts:

Dr. Jaume Bosch (jbosch@clinic.ub.es)
Dr. Fabio Piscaglia (fabio.piscaglia@unibo.es)
Dr. Ivan Amat-Roldán (ivan@ymaging.com)
Dra. Annalisa Berzigotti (Annalisa.Berzigotti@insel.ch)
Dra. Rosa Gilabert (gilabert@clinic.ub.es)