AI for complete pathological response evaluation with image-based machine learning models in esophageal carcinoma: AIFORCE
Suzanne Gisbertz, Dillen van der Aa, Henk Marquering, Jaap Stoker, Inez Verpalen, Jacqueline Bereska, Gert Meijer, Hanneke van Laarhoven, Roel Bennink, Annemieke Bartels-Rutten, Francine Voncken and the PRIDE-study group
Esophageal cancer is a prevalent malignancy worldwide with a poor prognosis, with 572.000 new patients and 508.600 esophageal cancer related death worldwide per year. Chemoradiotherapy (carboplatin, paclitaxel, and concurrent 41.4 Gy radiotherapy) for esophageal cancer followed by surgery is currently regarded as a standard of care in clinical practice in the Netherlands. A significant proportion of patients have a pathological complete response(pCR) following neoadjuvant chemoradiotherapy(nCRT). However, pre-operative determination via endoscopies and imaging remains unreliable for patient selection. The objective of this study is to develop an image-based machine learning prediction model to assess pathologic response (treatment effect) to nCRT in esophageal cancer, by evaluating 18F-FDG PET/CT and DW-MRI / DCE-MRI scans prior, during and after nCRT, compared to the histopathological assessment of the resection specimen (reference standard).
AI for improved prognosis and treatment response prediction using FDG PET/CT in multiple myeloma
Ronald Boellaard, Ben Zwezerijnen, Martijn Heijmans, Josée Zijlstra, PETRA consortium
Multiple myeloma (MM) is an aggressive malignancy of plasma cells in the bone marrow with an annual incidence of ~7 patients per 100,000 inhabitants. The disease cannot be cured and 20% of the patients do not respond (well) to standard therapy, resulting in an even worse outcome (progression free survival < 2 years). Better prognostification and selection of patients, as well as early response assessment based on (lack of) minimal residual disease will avoid futile treatments.
In this project we aim to apply, adapt and further develop AI methods to improve FDG PET/CT-based prognosis (for personalized treatment selection) using baseline scans and early response assessment after induction therapy based on minimal residual disease (MRD) for MM patients. The ultimate aim is to replace the difficult and variable visual reads and criteria by making more reliable, observer-independent quantitative AI reads of FDG PET/CT available and thereby better guide patient treatment selections and adaptions for clinical research.
The integration of deep learning for quantitative MRI in clinical and research workflows to personalize treatment of cervical and head and neck cancer
Myrte Wennen, Fellow & Oliver Gurney-Champion, Supervisor
Cancer patients can greatly benefit from precision medicine, where quantitative magnetic resonance imaging (MRI) biomarkers are used to select optimal personalized treatment. Unfortunately, the application of quantitative MRI faces challenges in accuracy, precision and reproducibility. Consequently, despite many promising results, quantitative MRI is not integrated in clinical care. Our research has shown that deep learning can drastically improve the accuracy and consistency of quantitative MRI techniques. At this moment, these deep learning techniques are not utilized efficiently, which greatly hinders implementation in clinical and research workflows. With this Fellowship, I aim to enhance precision medicine for cancer patients by integrating deep learning techniques for quantitative MRI directly on MR scanners and into clinical and research workflows. We will apply the techniques in two clinical trials investigating treatment response in patients with cervical and head and neck cancer.