1. To develop a predictive model that quantitatively estimates myocardial fi brosis using echocardiography in animal models and patients. This will be achieved through paired acquisitions of myocardial regions of interest (ROI), from which radiomics will be derived, and histological quantifi cation of percentage fi brosis in colocalised myocardial samples.
2. To assess the role of the echocardiography machine resolution and to compare image acquisitions from standard transthoracic echocardiography (phased-array probe) with epicardial myocardial acquisitions with a high resolution linear-probe in the performance of predictive models.
3. To appraise the added value of pulse-cancellation ultrasound (US) and US contrast administration with harmonic detection settings compared to standard echocardiography settings in these predictive models. This will be achieved by repeated image acquisitions under different conditions in the experimental models.
4. To test various alternatives of image preprocessing before radiomics feature extraction, thereby extending and optimizing the pipeline proposed by Kagiyama et al. [Kagiyama2020].
5. To develop and share open-source software tools that render semi-automated extraction ROIs from the left ventricle myocardium in images from standard echocardiography views.
6. To develop predictive models that quantitatively estimate myocardial fi brosis through echocardiography-derived radiomics features, using late gadolinium enhancement assessment and extracellular volume quantifi cation by myocardial longitudinal relaxation time mapping (T1 mapping) in cardiac magnetic resonance (CMR) imaging as a gold-standard reference. This objective will be achieved by comparing echocardiography and CMR image acquisitions in patients who underwent both modalities within a narrow time window by retrospectively querying the picture archiving and communication systems of two central hospitals.
7. To validate objective 6 and obtain model performance metrics in a test sample.
8. To make databases accessible online (ensuring anonymisation and ethical issues) as well as the whole data analytics pipeline.
9. To deploy models online and to desktop applications so that researchers and clinicians can upload new examinations and obtain real-time predictions. |
