ESTRO 2024 - Abstract Book

S4483

Physics - Machine learning models and clinical applications

ESTRO 2024

the model's ability to discern patients requiring varying levels of adaptive intervention, and gave scores of: Precision (0.85), recall (0.9), accuracy (0.91) and F1(0.88).

Conclusion:

The random forest decision tree model demonstrates promise in identifying patients who are likely to benefit from adaptive radiotherapy. The high number of false positives, initially perceived as inaccuracies, actually underscores the model's ability to detect patients in need of adaptive interventions, even when a complete rescan and retreatment scenario may not be warranted. This tool offers the potential for a proactive approach to adaptive radiotherapy in the local department, shifting from a reactive approach. The AI-based predictive model offers a more nuanced approach to patient selection, ensuring that those who require adaptive interventions, in varying degrees, are identified and treated accordingly. The transition from a reactive to a proactive approach has the potential to improve patient outcomes and streamline clinical practice in adaptive radiotherapy. This model holds promise for broader application in radiotherapy departments seeking to enhance patient selection and treatment planning.

Keywords: ART, Head and Neck, AI

1536

Digital Poster

MRI-based multi-head attention classifier for early detection of radiotherapy toxicity

Manish Kakar 1 , Olga Zlygosteva 2 , Inga Solgård Juvkam 1 , Nina Edin 2 , Eirik Malinen 1,2

1 Department of Radiation Biology, Institute for Cancer Research, The Norwegian Radium Hospital, Oslo University Hospital, Oslo, Norway. 2 Department of Physics, University of Oslo, Oslo, Norway

Purpose/Objective:

Radiotherapy (RT) of head and neck cancer may give toxicities that manifest both early and late [1]. If individuals at risk of toxicity can be identified at an early stage, one may opt to change the treatment or administer mitigating drugs. Non-invasive medical imaging such as magnetic resonance imaging holds a promising potential for monitoring changes in irradiated organs due to its high soft tissue contrast. Preclinical models with pertinent endpoints are ideal for accurately assessing normal tissue responses and their reliance on treatment parameters such as radiation field and type, dose, and fractionation scheme. The purpose of the study was to devise a classifier for early detection of radiation-induced toxicity in the head and neck region in mice using an MRI-based multi head attention classifier.

Material/Methods: