BINDER - Improving Bio-Inspired Deep Learning for Radiomics,

BINDER - Improving Bio-Inspired Deep Learning for Radiomics

According to the 2022 statistics from the American Cancer Society, breast and prostate cancers are the most prevalent worldwide, causing 670,000 and 35,250 deaths respectively. Axillary cancer, which may arise as a complication of breast cancer, has a 37% mortality rate. Several institutions, including the National Cancer Institute (NCI), have emphasized the urgent need for reliable predictive models to improve personalized therapies and forecast risk by incorporating relevant features.

Current predictive models often lack generalizability due to limited datasets from single institutions and patient cohorts, along with the limitations of existing AI algorithms. The NCI has called for substantial progress in Artificial Intelligence (AI) to address these issues.

To tackle this, MagIC coordinated the BINDER project, forming an interdisciplinary partnership to develop an AI-based system with unprecedented predictive ability in oncology. Utilizing novel and diverse data collected by CF, the project developed cutting-edge AI algorithms.

The BINDER project led to significant advancements in cancer diagnosis and treatment through two major contributions:

1.  a comparative study of breast cancer and axillary cancer, revealing previously unknown differences between these tumors;
2. an innovative study on the segmentation and detection of prostate cancer.

The BINDER project, coordinated by NOVA IMS, focused on developing an intelligent system to support clinical decision-making in oncology.

NOVA IMS began by analyzing the needs and demands of CF to improve their clinical practice, establishing a solid foundation for the project. Building on this understanding, NOVA IMS explored, designed, and examined optimal solutions, utilizing vast amounts of medical data, particularly images, and experimenting with innovative AI algorithms.

The development phase involved creating the AI-driven system, integrating diverse data types provided by CF, and employing advanced AI techniques. The system was then validated with CF clinicians and integrated into their existing framework, ensuring practical applicability.

Project coordination ensured effective collaboration and communication among stakeholders, while dissemination efforts shared the project's findings, promoting further innovation in predictive modeling in oncology.

The outcome of the BINDER project is a robust intelligent system that supports clinical decisions. CF now has access to several predictive models for breast and axillary cancer and an advanced system for the automatic detection and segmentation of prostate cancer. These models enhance early treatment, which is crucial for therapeutic success.

• Fundação Champalimaud
• Universidade de Coimbra
• Fciências.ID - Associação para a Investigação e Desenvolvimento de Ciências

• Funding Programme: FCT, I.P. - R&D Projects in All Scientific Domains - 2017
• Funding to NOVA IMS: 85.175€
• Duration: 2018 - 2022