AI for cancer identification and analysis
- Project title: Research and development of new AI-based software technology for the identification and analysis of cancer
- Acronym: VIT4CANCER
- File Number: INNCAD/2022/86
- Program: Consolidation of the business value chain
- Project Location: Paterna, Valencia (Spain)
- Start date: 05/06/2022 – End date: 04/30/2024?
- (partially) Funded by: GENERALITAT VALENCIANA: VALENCIAN AGENCY OF INNOVATION (AVI)
The project was born as a result of the limitations detected in the state of the art in the identification, analysis and assistance of cancer pre-diagnosis.
Research and development of new AI-based software technology for the identification and analysis of cancer through a novel distributed system based on image classification services in the cloud and client applications that help doctors in making cancer pre-diagnosis decisions in Health Centers for both Primary Care and Specialized Care. For this, different studies, research and developments will be carried out focused on the use of AI for the analysis and classification of images. With this, this project is associated with a variety of specific objectives focused both on the AI-based platform that will carry out this identification, as well as on the novelties and modifications that must be implemented in the current specialized platform called Adsum++, for this type of service. Adsum++ is IDI EIKON’s leading solution in the sanitary sector, focused on traceability and improvement of disease monitoring protocols.
Specific objectives associated with the project
- Study, analysis and comparison between different architectures and learning techniques that facilitate the generalization of the models (learning transfer, data augmentation, dropout, regularization, early stop, etc.).
- Research and development of a distributed system based on image classification services in the cloud and client applications that help physicians in making cancer pre-diagnosis decisions.
- Research, development, training and evaluation of deep learning models, from different types of images potentially available in Health Centers (RGB and ultrasound, …), based on different architectures of the “Vision Transformers” (ViT) type, comparing them with the “Convolutional Neural Networks” (CNN).
- Design and development of an IA platform that allows the integration of different types of IA models.
- Design and development of a platform for exposing the results of these image classification services in the form of applications focused on facilitating the interoperability of the results.
- Validation of the best model obtained as a shared cloud service to test its potential clinical utility in prospective trials through clinician-friendly client applications.
- Introduction of tools for better interpretability and explainability of the results obtained by the Artificial Intelligence model for the clinical user.
Expected results and technology developed
- Deep learning models, based on different types of images potentially available in Health Centers (RGB and ultrasound, …), based on different architectures of the “Vision Transformers” (ViT) type.
- IA Platform: which will allow the integration of different types of IA modules that use the models of IA architectures based on ViT. A basic module with the predictive functionality or an extended module that will add explainability and interpretability functionality for the clinical user. This platform will be integrated into the current IDI EIKON disease monitoring protocol platform.
- Platform for disease monitoring protocols: designed and developed by IDI EIKON, currently used by the company’s clients in the health sector. The objective of the platform is to work from any browser and device with access to the Intranet/Internet, which allows the implementation of digitalized healthcare plans or healthcare management strategies within a healthcare organization (by care area, by pathology, by patient profile…), transform each healthcare plan or healthcare route into an interactive calendar of tasks, measure the performance of each professional, team and center in the healthcare management of each healthcare plan, use the entire set of registered information to prepare different Dashboards and reports that allow the evaluation of the Clinical and Healthcare status of a patient or a cohort of patients, facilitating decision-making in daily clinical practice and also in the way of managing related health processes.