New PREDICT-Meso team member

We are delighted to welcome our latest team member, PhD student Josh Roche.


Radiological measurement of treatment response in mesothelioma is extremely difficult. The complex shape of the primary tumour makes surrogates of true volume unreliable – including modified RECIST criteria, which constitute the current clinical gold standard.  Measuring volume directly is too laborious for human readers.  PREDICT-Meso members have created and recently reported a prototype AI system, which can detect and measure Mesothelioma tumours on CT scans without any human input. The system uses deep learning, allied with extremely high-fidelity ground truth, and performed well in a retrospective multicentre cohort study of 183 CT datasets.

This project, which is part of the PREDICT-Meso International Accelerator Network will collate and utilise an expanded dataset of 2000 CT scans collected via the National Consortium for Intelligent Medical Imaging (NCIMI). The project will further optimise the AI developed and better calibrate volumetric response thresholds suitable for clinical use.  Once deployed into clinical practice, the AI is expected to improve clinical decision making and reduce future clinical trial costs. The successful completion of the project will also provide a strong proof-of-principle for development of similar tools for the assessment of other cancers.

Research questions

  • Can AI performance be further improved by inclusion of cases with challenging morphological features (e.g. fissural tumour, adjacent atelectasis, contralateral pleural disease) in future training sets?
  • Can AI volumetric response thresholds be better calibrated to therapeutic behaviours?

Welcome to the team Josh!