Student Name: YAQEEN ALI
Summary:
Yaqeen Ali has been working as a research associate at Mediri GmbH in Heidelberg, Germany, since March 1, 2023. He is doing work for the BosomShield project, and his main goal is to develop a cloud-based CAD prototype for BosomShield. As well, he enrolled as a PhD student under the supervision of Prof. Dr. Johannes Gregori and Prof. Andreas Weinmann at the Doctoral Centre for Applied Computer Sciences at the Damstardt University of Applied Sciences. Yaqeen Ali is originally from Sargodha, Pakistan. He has experience with deep learning and computer vision in medical imaging. Before joining the BosomShield project at Mediri, he had completed his master degree in computer science (MSCS) in July 2022 from Comsats University Islamabad, Lahore Campus, Pakistan. During his master thesis, he worked on automated white matter hyperintensity segmentation on brain MR images with deep learning. Before this, he earned his bachelor of science in computer science (BSCS) from the University of Sargodha. In close collaboration with the Interdisciplinary Research Centre in Biomedical Materials (IRCBM) at Comsats University Islamabad, Lahore, he worked on the prediction of acetone from a portable paper-based filter using machine learning techniques for his becholor project.
Objectives:
Mediri’s software solution mTRIAL already contains image upload, de-identification and processing functionality for DICOM standard images, especially mammograms, MRI, ultrasound, 3D ultrasound and WSI. The main objective is to develop a CAD system prototype that combines decentralized support of FL models for multi-modal image processing with a secure cloud-based image analysis platform, which will be easily accessible via a web browser while preserving data privacy. For the productive phase of the cloud-based CAD prototype, patient data will be pseudonymised and encrypted before transfer to the site-specific secure server location, where the XAI models are performed. A UI based co-creation approach combines the results with the local patient information again to support the local clinician. Mediri has filed a patent on this inherent data privacy design. The scientific objectives of this DC are 1) to develop, investigate and evaluate an FL concept targeted to the specific projects need, combined with the cloud system. 2) Commonly used DL models will be integrated into the workflow of the system, developing iteratively the integration of a modality independent learning pipeline. 3) Common architectural concepts favour stand-alone software solutions connected to internal clinical IT networks (PACS) and a secure web-based CAD system with an inherent data privacy design. This approach is sensible looking at the large size of data in multi-modal breast image analysis, which can reach several GB per dataset, and the demand of specialised user interactions. Alternatively, cloud-based architecture approaches gain increasing relevance in XAI. The advantages are to have a high requirement on computing power by easily scalable parallel cloud-based computing methods and to easily manage scalable data from external sites in parallel, taking the advantage of the outcomes of the models developed through the FL system.
Outcomes:
1) CAD prototype software with cloud-based image data management for MRI, ultrasound, 2D/3D mammography, and biopsy images among other. 2) Prototype of interfaces to publicly available commercial cloud computing resources. 3) Integration or interfaces to automated multi-modal image processing algorithms developed in WP4, WP5 and WP6 and trained using FL techniques (DC9).