Multifactorial decision support for the treatment of malignant melanoma 

Skin cancer encompasses three main types: melanoma, basal cell carcinoma and squamous cell carcinoma. Malignant melanoma accounts for only 2% of skin cancers but is responsible for 75% of skin cancer deaths. It is a very aggressive disease that can typically metastasize anywhere in the body. Thanks to the advances in immunotherapy, the 5-year survival rate for metastatic melanoma has increased from less than 5% to 30%. However, it is still impossible to predict who will respond to the treatment. The goal of this project is to develop a clinical decision support system for malignant melanoma to support the medical experts in the challenging treatment of this disease.

Positron emission tomography / computed tomography (PET/CT) is an imaging modality that is widely used in oncology because it offers the combination of functional and anatomical information. For the PET, a radioactive tracer is used, in this case fluor-18-fluordeoxyglucose (18F-FDG). This is a glucose analogue, so all areas with metabolic activity, in which glucose is used, will light up on the PET scan. This includes metabolically-active tumours, but also other regions that consume glucose, like the brain and certain parts of the abdomen. Latter regions will light up the same way as a lesion but are due to normal physiological tracer uptake. This way, PET offers the functional information with high sensitivity and specificity. The CT offers anatomical information with a high resolution and is used to get an idea of the exact location of the areas that light up on the PET scan.

"My ultimate goal is to offer a better outlook to people diagnosed with melanoma."

Because malignant melanoma can metastasize anywhere in the body, whole-body imaging is necessary. However, these images still pose several challenges in the clinic, including the large amount of data for manual reading, the varying appearances of lesions across the body and cumbersome reporting. To assess the state and progression of the disease, several quantitative parameters have been reported to be of value, including imaging-based measures such as the total metabolic tumour volume and total lesion glycolysis. Other indicators, like the presence of brain metastases and certain blood values at baseline, are vital for optimal treatment selection. Accurately localizing and delineating all the positive lesions from whole-body acquisitions, required for all of the above, is a time-consuming and error-prone task. Current software packages can aid the nuclear physician and radiologist in this task, yet many shortcomings remain to be overcome, and automated analysis is currently not feasible, hampering analysis of a large number of patients. Data science can play a complementary role to clinical studies thanks to automated reading of a high number of scans, automated extraction of features, the ability to assess a wide range of features, etc.

In consultation with the medical experts, three important clinical questions are formulated that could improve the disease management. The first one is to predict if the patient will respond to the standard treatment at the moment the first scan is taken. If at this stage, it can already be confirmed that the standard treatment will have no effect, a more experimental therapy can be started much earlier, offering the patient a better chance of a positive outcome. The second clinical question to address is to predict the response to treatment during the initial follow-up. This way the treatment can still be altered if it doesn’t give satisfying results. The final question is when it is safe to stop the treatment. Currently, for patients with good response, treatment is halted based on subjective criteria and considering the patients wishes. While killing all cancer cells is important, not needlessly burdening the patient with a heavy therapy is also. In addition, reduction of treatment schedules, without loss of efficacy, may allow other patient groups to benefit from this costly treatment approach. The improved stratification of patients provided by the system will allow better allocation of expensive resources such as immunotherapy while avoiding toxic, unnecessary treatments for others.

This project aims to develop a clinical decision support system including automated segmentation tools for whole-body PET-CT, the extraction of prognostic and predictive features from imaging, and the development of predictive models for treatment outcome and response in case of malignant melanoma.

Achievements (Honors & Awards) 
  • 2018: ie-net press prize
  • 2018: ie-net 3th place in the category civil engineering