Aid System for the Detection of Various Pathologies Based on Image Analysis and Artificial Intelligence
The CADIA project aims to create an image analysis system based on state-of-the-art artificial intelligence techniques (e.g. deep learning). As clinical scenario for the piloting of the above mentioned system it has been considered the help to the screening of breast cancer, developing algorithms for the analysis of the lesions in images of mammography and digital tomosynthesis, and diagnosis of breast cancer, by means of new techniques of help to the detection and characterization of tumor zones in images of digital pathology.
Challenges: In radiology, to design a CAD system capable of detecting masses in mammography and digital tomosynthesis images acquired with all the equipment available in SERGAS, so that all patients, regardless of the hospital they are in, have access to the system. For pathology, the scenario is more novel and requires the jump to digital pathology and the analysis of these digitized images for the diagnosis of the type of lesion and its degree. In addition, any deployment infrastructure must be developed in an integrated manner with the health service's own systems.
Vicomtech leads the UTE, therefore, as leader and role, it is in charge of the coordination and collection of requirements and specifications together with others. Besides, it is the main responsible for the development of the AI algorithms and the necessary infrastructures for the creation of models and inference. He has also been largely responsible for the creation of the database, including the development of annotation tools.
The project is being developed by the joint venture formed by the Vicomtech Foundation and the company Inycom. The present Innovative Public Procurement project is framed within the health innovation plan Code 100 executed in the framework of a collaboration agreement between the Galician Health Service and the Ministry of Science and Innovation, financed in an 80% by the ERDF in the framework of the Spanish Multiregional Operational Programme (POPE) 2014-202.
Currently, the necessary annotation tools have been created with which the clinicians are labeling the images, after long sessions of gathering requirements and specifications, as well as training and annotation tests. 2400 pathological anatomy samples have been digitized, creating a database of anonymized dicom images and the necessary sergas PACS radiology images have been selected and anonymized. The first AI models from public digital radiology and pathology databases have been trained.