The Rheumatic Heart Disease (RHD) can be easily treated in its early stages. However, it may produce severe damage to the heart if it remains untreated, including severe sequelae and death.
Thanks to this use-case, stakeholders will have valuable clinical information obtained through a novel technological approach. Moreover, its integration within the ATMOSPHERE environment will allow to assess the robustness and validity of the platform for the future.
Clinical features will be extracted from Doppler ultrasound images by using a radiomics approach which includes pre-processing and analysis algorithms, such as artefact removal, segmentation and texture analysis. These features will be analysed using data reduction and classification techniques in order to find significant relationships with the clinical endpoints of the RHD.
This strategy will be accompanied by deep learning techniques based on Convolutional Neural Networks (CNN) to create a classifier for the screening of echo-cardio images on this pathology. The classifier will facilitate the differentiation between normal and abnormal (definite and borderline RHD) studies.
The present method represents a new approach to handle the RHD. Its implementation will help physicians with quantitative information to achieve an early diagnosis, thus reducing the workload and improving the time-efficiency of health personnel while enhancing patient outcomes.
Design of a 3D Convolutional Neural Network as feature extractor.