Delta-radiomics data science evaluation, validation and integration into the precision oncology pathway for bone and soft tissue sarcoma.

Background: High-grade sarcomas are lethal cancers, with a high risk of recurrence (10-20%), metastatic disease (30-40%) and death. Personalised treatment offers potential to improve sarcoma outcomes.

This requires a more accurate description of the sarcoma at that time and knowledge of the features within the sarcoma that are characteristic of behaviour and response to treatment. Advanced medical image analysis contributes to both requirements.

Three-dimensional medical imaging modalities such as Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) are standard-of-care.

However, quantitative image information is not frequently used with current clinical image interpretation often limited to subjective visual assessment.

Extraction of additional quantitative information from longitudinal medical imaging data is termed 'delta-radiomics' and can provide clinicians with additional information about tumours and response to treatment.

Aim: This project will develop an automated analysis pipeline for delta radiomics that provides clinicians with additional information and assesses response to treatment. The overarching aim is increased personalisation of treatment and improved sarcoma outcomes.

Objectives and Timeline: Develop an automated analysis pipeline to apply existing delta-radiomic methodology to extract image features in a retrospective longitudinal cohort of ~1,000 high-grade sarcoma patients (~3,000 scans) [Year 1].

Develop, validate and test a delta-radiomic processing model for sarcoma to provide additional information on outcomes (recurrence, metastatic spread and response to treatment), including assessment of model robustness and reproducibility [Year 2].

Test automated analysis pipeline developed in the prospectively acquired Oxford Precision Oncology for Sarcoma (OxPOS) dataset [Year 3].

Assess feasibility of automatic analysis pipeline developed considering robustness, reproducibility, and overall pipeline performance [Year 3].

Methods: The research proposed has access to a large longitudinal retrospective Oxford Sarcoma Service imaging dataset (n~1,000) for development and validation of the automated pipeline and model. The pipeline and model developed will also be tested in a prospectively acquired dataset (n~250) from the OxPOS study.

Machine learning techniques will be used to learn which radiomic features are characteristic for outcomes (recurrence, metastatic spread and response to treatment).

Model performance will be assessed through measurement of precision, sensitivity, and specificity in validation and testing.

The robustness of the pipeline to changes in images will be assessed through patient and phantom images, the most robust features will be those that demonstrate the smallest variation with changes to imaging parameters.

Anticipated Impact and Dissemination: An advanced image analysis pipeline enabling the extraction of delta-radiomic features from standard-of-care imaging will be developed.

Through analysis and testing in a large retrospective dataset, and further testing in the prospectively acquired OxPOS dataset, delta-radiomic features will be used to provide information on response to treatment and sarcoma behaviour. The feasibility assessment will inform a future multi-centre study looking at a wider patient population.

A PPI advisory group will be engaged throughout the project. Confidence in the adoption of new technology is vital to successful implementation. The PPI group will contribute to the production of visual and written lay summaries for effective communication. Results will also be disseminated at relevant scientific conferences and in peer-reviewed scientific publications.