Artificial Intelligence in Medicine: focus on Multi-Input Analysis
The project focuses on a main methodological open issue related to the development of AI-based tools for medical data analysis: 1) mining multi-input data, i.e. providing new analysis strategies to exploit complementary information encoded in data coming from heterogeneous sources.
Making progress in this field also requires addressing the following key co-occurring challenges: 2) handling incomplete/missing/limited datasets; 3) developing a dedicated data and IT platform for secure data management, linked to adequate computing resources.
To achieve these goals, sharing data and knowledge within a broad scientific community (networking) is a key element.
The investigation of different diagnostic tests and clinical information available for a patient allows medical experts to define an integrated clinical profile to outline the most appropriate care path.
This task could be supported by AI, through the design of data analysis models that accept heterogeneous input, thus allowing multidimensional analysis.
It is not straightforward to develop models capable of analyzing digital data acquired with different instruments, given their high variability in format, dimensionality and informative content.
The objectives of the project will be achieved by means of the realization of four interconnected WPs:
WP1. Mining multi-modal information. AI-based analysis pipelines to handle and mine a combination of heterogeneous data sources, which may include medical images, diagnostic tests, phenotypic and genetic data will be developed and validated.
WP2. Handling incomplete/missing/limited datasets. Computational strategies for data curation, imputation and augmentation will be developed to avoid severe reductions of the sample, and to compensate for unbalanced or small samples.
WP3. Data platform and computing infrastructure. To achieve the AIM_MIA’s goals, large multi-modal data samples are needed. To ensure effective access to data shared within the collaboration, including raw data and processed information, a dedicated data platform integrated with computing resources, and compliant with the FAIR principles and the GDPR regulation will be developed.
WP4. Project management, outreach and networking. This WP deals with activity planning and monitoring of project progression, fostering the cooperation and exchange of ideas among researchers from the different groups involved.