Periodic Reporting for period 1 - RadioVal (International Clinical Validation of Radiomics Artificial Intelligence for Breast Cancer Treatment Planning)
Reporting period: 2022-09-01 to 2024-02-29
Breast cancer, the most prevalent cancer globally, accounts for nearly 30% of all cancers in women, with its incidence on the rise. Developed nations report higher breast cancer incidence rates, while lower-income countries face elevated mortality rates due to cost barriers and limited treatment access, potentially resulting in under-treatment.
Neoadjuvant chemotherapy (NAC), a promising treatment for advanced breast cancer, has demonstrated efficacy in improving prognosis. However, not all patients respond effectively to NAC, and its associated side-effects vary, affecting its overall effectiveness. To address these challenges, personalised approaches are needed to assess individual responses, side-effects, and prognosis to improve decision-making. Artificial intelligence (AI), particularly radiomics-based algorithms, hold potential in predicting patient responses to NAC, aiding in more tailored treatment decisions and reducing both over- and under-treatment.
The main objectives of RadioVal are:
1. Implement the very first international, multi-faceted clinical validation study for radiomics-based prediction of response to NAC in multiple developed and developing countries.
2. Introduce a holistic, standardised methodological framework for multi-faceted and trustworthy evaluation of radiomics AI, taking into account multiple technical, clinical as well as ethical criteria.
3. Implement a multi-stakeholder, inclusive approach to improve awareness, acceptance and promotion of radiomics AI in future breast cancer care.
4. Develop the very first traceability tool for radiomics AI, which will enable transparent monitoring and continuous evaluation of radiomics tools during their lifetime.
5. Evaluate wider impacts of clinical deployment of radiomics AI, including associated cost-benefits, socio-ethical implications and regulatory aspects.
Neoadjuvant chemotherapy (NAC), a promising treatment for advanced breast cancer, has demonstrated efficacy in improving prognosis. However, not all patients respond effectively to NAC, and its associated side-effects vary, affecting its overall effectiveness. To address these challenges, personalised approaches are needed to assess individual responses, side-effects, and prognosis to improve decision-making. Artificial intelligence (AI), particularly radiomics-based algorithms, hold potential in predicting patient responses to NAC, aiding in more tailored treatment decisions and reducing both over- and under-treatment.
The main objectives of RadioVal are:
1. Implement the very first international, multi-faceted clinical validation study for radiomics-based prediction of response to NAC in multiple developed and developing countries.
2. Introduce a holistic, standardised methodological framework for multi-faceted and trustworthy evaluation of radiomics AI, taking into account multiple technical, clinical as well as ethical criteria.
3. Implement a multi-stakeholder, inclusive approach to improve awareness, acceptance and promotion of radiomics AI in future breast cancer care.
4. Develop the very first traceability tool for radiomics AI, which will enable transparent monitoring and continuous evaluation of radiomics tools during their lifetime.
5. Evaluate wider impacts of clinical deployment of radiomics AI, including associated cost-benefits, socio-ethical implications and regulatory aspects.
The RadioVal project has made significant advancements towards achieving its scientific objectives. Clinical and technical partners have collaborated to define and collect appropriate clinical data for the model development and validation stage. While the data is being collected, more than 1,500 multi-vendor, multi-centre NAC cases have been harmonised from public datasets, enabling us to begin fine-tuning AI algorithms and setting up the pipeline for validation at clinical sites. This data will be combined with data from the RadioVal consortium (and beyond) to produce reliable prediction models for NAC treatment.
A consensus list of metrics for multi-faceted Radiomics evaluation has been developed, along with a draft quantitative framework for AI model evaluation. AI models are being developed in alignment with FUTURE-AI principles to enhance trustworthiness and clinical acceptance of such tools. Methods for evaluating Radiomics cost-effectiveness have been explored, and case selection for Radiomics models within RadioVal has been completed, considering diverse real-world clinical data.
A full set of relevant stakeholders has been identified, and their interests and influence towards certain aspects of the RadioVal solution have been analysed. The methodology for stakeholder inclusion has been defined, and engagement activities have commenced with three rounds of social innovation sessions. The preliminary version of the AI passport, designed as a traceability tool, will undergo testing by consortium members. Image evaluation quality control is complete, while segmentation quality control and the design of a human-in-the-loop mechanism for clinical feedback have begun and will be finalised in the coming months.
A consensus list of metrics for multi-faceted Radiomics evaluation has been developed, along with a draft quantitative framework for AI model evaluation. AI models are being developed in alignment with FUTURE-AI principles to enhance trustworthiness and clinical acceptance of such tools. Methods for evaluating Radiomics cost-effectiveness have been explored, and case selection for Radiomics models within RadioVal has been completed, considering diverse real-world clinical data.
A full set of relevant stakeholders has been identified, and their interests and influence towards certain aspects of the RadioVal solution have been analysed. The methodology for stakeholder inclusion has been defined, and engagement activities have commenced with three rounds of social innovation sessions. The preliminary version of the AI passport, designed as a traceability tool, will undergo testing by consortium members. Image evaluation quality control is complete, while segmentation quality control and the design of a human-in-the-loop mechanism for clinical feedback have begun and will be finalised in the coming months.
RadioVal will implement a comprehensive clinical evaluation of radiomics-driven estimation of NAC response, involving a diverse sample of EU training cases and multi-continent testing cases through the AI4HI Network. Three key innovations will drive this initiative. Firstly, a novel radiomics evaluation framework will be established, incorporating specific criteria, metrics, and methods like the Radiomics Usability Scale and specialised cost-effectiveness analysis. Moreover, transfer learning and domain adaptation will be employed to evaluate the applicability of radiomics solutions across Europe and beyond. Secondly, a radiomics traceability tool will be developed, including a radiomics passport for model monitoring, along with functions for quality control, failure detection, periodic evaluation, and continuous learning. Finally, a comprehensive multi-stakeholder radiomics information and communication package will enhance awareness and inclusion in radiomics-driven prediction of NAC response in breast cancer. Moreover, RadioVal will establish a legal governance framework to enable data sharing while upholding compliance with data privacy regulations and safeguarding citizen rights. Additionally, the project will devise standards and best practices to ensure future AI solutions are technically robust, clinically safe, and devoid of bias. The multi-stakeholder approach will foster shared decision-making and dialogue between clinicians and patients/caregivers, improving tool effectiveness and patient engagement. Social innovation will guide innovations and evaluations, considering diverse needs and contexts, while systematically analysing clinical, ethical, and regulatory implications. To ensure further uptake and success, the project has identified crucial requirements including follow-up research, end user testing, real-world demonstrations, and legal counsel support.