Periodic Reporting for period 2 - PRE-ACT (PRE-ACT: Prediction of Radiotherapy side Effects using explainable AI for patient Communication and Treatment modification)
Période du rapport: 2024-04-01 au 2025-09-30
The PRE-ACT project delivers a framework, grounded on solid and novel human-interpretable AI concepts, to predict the risk of side effects of radiotherapy treatment for breast cancer patients and subsequently utilize it to inform the patients about optimal treatment. The objectives are:
•Leverage data from 3 multi-centre patient European cohorts to train AI models for risk prediction of the occurrence of side effects with a primary focus on arm lymphedema.
•Homogenise and analyse data consisting of various modalities and include patient medical records such as comorbidities, anatomy, demographics, as well as treatment data, radiotherapy dose distribution data, Computerized Tomography (CT) scans, auto-contouring of critical organs in CT scans, and genetic data.
•Utilize advanced explainable AI (XAI) models that provide explanations about the risk of developing arm lymphoedema and other side effects, and study the transferability of models to other types of cancers such as prostate cancer.
•Create an actual testbed within the controlled environment in the AUEB-RC' lab to implement and deploy various FL algorithms to simulate real-world scenarios of data owners around the world that would like to collaboratively train AI models while keeping their data separate.
•Utilize Federated Learning as a proof of concept to assess the quality of predictions when the data are private and decentralized.
•Develop a dose prediction mode.
•Assess the impact of explainability of the AI model in a clinical investigation that comprises two arms, namely two disjoint subsets of recruited patients. In the first arm, the personalized risk prediction will be communicated to physicians and patients, while in the second arm, it will not.
•Assess the impact of communicating a personalized prediction of lymphoedema risk and prescribing a prophylactic arm sleeve (in case of elevated risk), on the occurrence of the arm lymphedema, the radiation treatment planning and the patients’ quality of life.
•Adopt a participatory co-creation and co-design process with involved stakeholders (patients, physicians, radiation oncologists) and consider the process from the user requirements elicitation to the assessment of the communication package.
•Design and build user-friendly app platform to present risk predictions and explanations in a clear and understandable way and facilitate communication and collaboration between patients and doctors.
•Address fairness considerations in the AI algorithms that aim at uncovering and explaining potential biases in data and provide explanations about their nature.
•Devise probabilistic and optimization models to study the health economics dimension and potentially reduce healthcare costs.
•Leverage data from 3 multi-centre patient European cohorts to train AI models for risk prediction of the occurrence of side effects with a primary focus on arm lymphedema.
•Homogenise and analyse data consisting of various modalities and include patient medical records such as comorbidities, anatomy, demographics, as well as treatment data, radiotherapy dose distribution data, Computerized Tomography (CT) scans, auto-contouring of critical organs in CT scans, and genetic data.
•Utilize advanced explainable AI (XAI) models that provide explanations about the risk of developing arm lymphoedema and other side effects, and study the transferability of models to other types of cancers such as prostate cancer.
•Create an actual testbed within the controlled environment in the AUEB-RC' lab to implement and deploy various FL algorithms to simulate real-world scenarios of data owners around the world that would like to collaboratively train AI models while keeping their data separate.
•Utilize Federated Learning as a proof of concept to assess the quality of predictions when the data are private and decentralized.
•Develop a dose prediction mode.
•Assess the impact of explainability of the AI model in a clinical investigation that comprises two arms, namely two disjoint subsets of recruited patients. In the first arm, the personalized risk prediction will be communicated to physicians and patients, while in the second arm, it will not.
•Assess the impact of communicating a personalized prediction of lymphoedema risk and prescribing a prophylactic arm sleeve (in case of elevated risk), on the occurrence of the arm lymphedema, the radiation treatment planning and the patients’ quality of life.
•Adopt a participatory co-creation and co-design process with involved stakeholders (patients, physicians, radiation oncologists) and consider the process from the user requirements elicitation to the assessment of the communication package.
•Design and build user-friendly app platform to present risk predictions and explanations in a clear and understandable way and facilitate communication and collaboration between patients and doctors.
•Address fairness considerations in the AI algorithms that aim at uncovering and explaining potential biases in data and provide explanations about their nature.
•Devise probabilistic and optimization models to study the health economics dimension and potentially reduce healthcare costs.
PRE-ACT is progressing well across all activities, including the central workflow which culminates in the PRE-ACT-01 clinical investigation and the individual technical objectives of each WP. The main achievements of PRE-ACT in the first 36 months are:
1. Unify data from 3 different breast cancer radiotherapy cohorts into a central database and FAIR-ify them with the RDF framework
2. Re-analyze the CT scans from all patients to mark the organs-at-risk in a consistent manner across the cohorts.
3. Carry out a genetic analysis of SNPs that might predict arm lymphedema
4. Use ML to build predictive models for arm lymphedema, evaluating clinical, genetic, dosiomic and radiomic features using a range of ML approaches, calculating predictive accuracy across the whole patient population and in defined sub-sets e.g. patients with lymph node involvement
5. Develop Fidex and FidexGlo for local and global explainability, optimizing fidelity and sample coverage, and the Discretized Interpretable Multilayer Perceptron (DIMLP) ensemble model (with AUC 0.832) for clinical use
6. Develop a working prototype dose-prediction tool based on a prescription of 40 Gy over 15 fractions
7. Develop the PF-MOL framework to optimize the trade-off between model accuracy and explainability and improve fidelity (up to 99%) and F1 score (up to 102%); and the CFSL method that reduced training time up to 50%
8. Develop a clinical investigation proposal for validation of the AI prediction, and launch the PRE-ACT-01 trial in France in September 2025, targeting 724 patients across 35 sites (after securing regulatory approval, and deploying the trial infrastructure).
9. Get ethical approval granted for the pivotal stage clinical investigation of the medical device XAINET in France; 8 site initiation visits were completed, and 6 sites have been activated.
10. Develop an added-value, ancillary study to the PRE-ACT-01 study which will enable use of the patient cohort in future studies of genomic and transcriptomic predictors of radiotherapy toxicity
11. Design the PRE-ACTOR web app to inform patients and doctors of the individual risk for arm lymphedema after a participatory phase with key stakeholders
12. Build the PRE-ACTOR web app with explainable AI (XAI) modules, personalized recommendations, and risk visualizations and perform usability testing with 13 clinicians that confirmed clarity and ease of use
13. Identify sensitive socioeconomic and demographic attributes that may influence fairness and lead to bias in the AI models, such as income and education level
14. Develop a new framework, Size-adaptive hypothesis testing for fairness - SAFT to ensure fairness testing is statistically sound and size-adaptive
15. Develop a probabilistic model comparing current economics benefits that shows minor cost increases and improved Quality Adjusted Life Years (QALYs).
1. Unify data from 3 different breast cancer radiotherapy cohorts into a central database and FAIR-ify them with the RDF framework
2. Re-analyze the CT scans from all patients to mark the organs-at-risk in a consistent manner across the cohorts.
3. Carry out a genetic analysis of SNPs that might predict arm lymphedema
4. Use ML to build predictive models for arm lymphedema, evaluating clinical, genetic, dosiomic and radiomic features using a range of ML approaches, calculating predictive accuracy across the whole patient population and in defined sub-sets e.g. patients with lymph node involvement
5. Develop Fidex and FidexGlo for local and global explainability, optimizing fidelity and sample coverage, and the Discretized Interpretable Multilayer Perceptron (DIMLP) ensemble model (with AUC 0.832) for clinical use
6. Develop a working prototype dose-prediction tool based on a prescription of 40 Gy over 15 fractions
7. Develop the PF-MOL framework to optimize the trade-off between model accuracy and explainability and improve fidelity (up to 99%) and F1 score (up to 102%); and the CFSL method that reduced training time up to 50%
8. Develop a clinical investigation proposal for validation of the AI prediction, and launch the PRE-ACT-01 trial in France in September 2025, targeting 724 patients across 35 sites (after securing regulatory approval, and deploying the trial infrastructure).
9. Get ethical approval granted for the pivotal stage clinical investigation of the medical device XAINET in France; 8 site initiation visits were completed, and 6 sites have been activated.
10. Develop an added-value, ancillary study to the PRE-ACT-01 study which will enable use of the patient cohort in future studies of genomic and transcriptomic predictors of radiotherapy toxicity
11. Design the PRE-ACTOR web app to inform patients and doctors of the individual risk for arm lymphedema after a participatory phase with key stakeholders
12. Build the PRE-ACTOR web app with explainable AI (XAI) modules, personalized recommendations, and risk visualizations and perform usability testing with 13 clinicians that confirmed clarity and ease of use
13. Identify sensitive socioeconomic and demographic attributes that may influence fairness and lead to bias in the AI models, such as income and education level
14. Develop a new framework, Size-adaptive hypothesis testing for fairness - SAFT to ensure fairness testing is statistically sound and size-adaptive
15. Develop a probabilistic model comparing current economics benefits that shows minor cost increases and improved Quality Adjusted Life Years (QALYs).
A new global rule extraction algorithm (FidexGlo) that generates propositional rules from several ML models, including CNNs.
The use of patient and treatment characteristics in a predictive model to identify individuals at high risk of developing arm lymphedema.
A new dose prediction model
A novel Multi-objective optimization framework for the dual problem of improving accuracy and explainability in the context of FL
The Size-Adaptive Fairness Testing (SAFT), a new hypothesis-testing framework that adapts to subgroup size and establishes resolution limits for reliable decisions.
The project innovations—the FAIR-compliant data pipelines, explainable AI models, methods for assessing model fairness, the dose prediction model and the Federated Learning frameworks—are inherently generalizable. They provide a blueprint for deploying trustworthy AI in diverse healthcare contexts, ensuring interoperability, fairness, and cost-effectiveness across multiple cancer types and treatment modalities.
The use of patient and treatment characteristics in a predictive model to identify individuals at high risk of developing arm lymphedema.
A new dose prediction model
A novel Multi-objective optimization framework for the dual problem of improving accuracy and explainability in the context of FL
The Size-Adaptive Fairness Testing (SAFT), a new hypothesis-testing framework that adapts to subgroup size and establishes resolution limits for reliable decisions.
The project innovations—the FAIR-compliant data pipelines, explainable AI models, methods for assessing model fairness, the dose prediction model and the Federated Learning frameworks—are inherently generalizable. They provide a blueprint for deploying trustworthy AI in diverse healthcare contexts, ensuring interoperability, fairness, and cost-effectiveness across multiple cancer types and treatment modalities.