Verslagen

Samenvatting van de aanvraag

Samenvatting
Dit item is dichtgeklapt
Dit item is opengeklapt

Research question

Immunotherapy is a great advancement for patients with metastatic cancers. Approximately 60% of patients with metastatic melanoma receive immunotherapy. However, immunotherapy is very expensive, and ineffective in over 40% of the patients. Furthermore, it takes at least 5 months to ascertain non-response, wasting valuable time and money on ineffective treatment, and exposing patients to severe side-effects.

 

Hypothesis

We propose a research program that aims to develop deep learning based tools for pre-treatment response prediction in patients with metastatic melanoma.

 

Study design

Building on our experience with advanced image analysis and deep learning, we aim to unlock the predictive information from diagnostic modalities that are already part of routine clinical melanoma care: computed tomography scans and digital pathology slides.

The research will be carried out by a multi-disciplinary team of experts in pathology, radiology, oncology, radiotherapy and artificial intelligence, as well as cost-effectiveness. We will collaborate with existing biobanking initiatives aimed at understanding how tumors interact with the immune system.

 

Study population

With computerized image analysis and deep learning applied to these images, we will re-analyze existing data from approximately 500 melanoma patients treated with immunotherapy in The Netherlands. Furthermore, we will validate the predictive power of our models in a multi-center prospective cohort of over 750 patients.

 

Intervention

Patients will receive usual care and there will be no intervention based on study results.

 

Outcome measure

Best overall response by RECIST and irRECIST.

 

Sample size/data analysis

The 95% confidence interval around the negative predictive value of 90% will be 84.6%-94.0% with a prospective sample size of 750. Neural networks will be designed to make predictions primarily using radiology and pathology images as input, eventually allowing for inclusion of clinicopathological data.

Naar boven
Direct naar: InhoudDirect naar: NavigatieDirect naar: Onderkant website