Malignant melanoma is the most lethal form of skin cancer with a drastic drop in survival chances once it’s metastasized. Historically, five-year overall survival chances were around 10% or even less, depending on the location of the metastases. However, thanks to advances in the treatment options, survival chances have increased to around 30%. Though this is a great improvement, the majority of the patients will still fail to respond. Personalised parameters that are related to a patient’s survival under a certain therapy, could allow a more beneficial treatment selection and eventually increase the overall survival chances. Clinical research has identified a number of biomarkers, some derived from medical images, to be prognostic. Medical imaging with fluorine-18 fluorodeoxyglucose ([18F]FDG) positron emission tomography / computed tomography (PET/CT) offers valuable information needed for accurate diagnosis, response prediction and follow-up. Because melanoma can metastasize anywhere in the body, whole-body imaging is essential for proper assessment of disease status. However, the identification and quantification of lesions require time-consuming and labour-intensive manual work.

This is inherently prone to errors and subject to intra- and interreader variability, yielding low reproducibility. Automated image analysis can play a pivotal role in providing (potentially) prognostic imaging features to be exploited and researched.

Image-derived parameters on the status of the disease, like total metabolic tumour volume (TMTV), have shown their value in survival prediction [1-4]. Imaging features related the physical state of the patient, like body composition, have proven useful in other pathologies [5-7] but remain to be investigated for melanoma.

The objective of this proposal is to use image processing and machine learning for aiding the estimation of a patient’s survival chances under a certain treatment. The project will comprise two main steps. Firstly, promising features will be extracted from the whole- body PET/CT images. Secondly, available and newly extracted parameters will be combined in predictive models for survival.

The developments will be performed as an extension to existing software developed within the ETRO research group. The algorithms will be implemented in Python, using open-source image processing and machine learning.
The project will involve:

• Literature study.


• Implementation of image processing methods for feature extraction.


• Implementation of machine learning methods for predicting survival.


• Training and application of the tools for different patient sequences. A dataset of
  69 patients treated at UZ Brussel is available. We will investigate the RIC-MEL
  dataset [8] for external validation.


• Thesis writing.

References
[1] G. Awada, I. Özdemir, J. Schwarze, et al. . “Baseline total metabolic tumor volume assessed by 18FDG-PET/CT predicts outcome in advanced melanoma patients treated with pembrolizumab”. Annals of Oncology, 29(supplement 10, X7), 2018. doi: https://doi.org/10.1093/annonc/mdy493.019.
[2] G. Awada, J. K. Schwarze, O. Gondry, et al. . “Baseline biomarkers correlated with outcome in advanced melanoma treated with pembrolizumab monotherapy”. Journal of
Clinical Oncology, 38(15), 2020. doi: https://doi.org/10.1200/JCO.2020.38.15_suppl.e22041.
[3] G. Awada, Y. Jansen, J. Schwarze, et al. . “A comprehensive analysis of baseline clinical characteristics and biomarkers associated with outcome in advanced melanoma patients treated with pembrolizumab”. Cancers, 13(2):1–18, 2021. doi: https://doi.org/10.3390/cancers13020168.
[4] I. Dirks, M. Keyaerts, B. Neyns, I. Dirven, and J. Vandemeulebroucke. "Development and Validation of a Predictive Model for Metastatic Melanoma Patients Treated with Pembrolizumab Based on Automated Analysis of Whole-Body [18F]FDG PET/CT Imaging and Clinical Features". Cancers, 15(16):4083, 2023. doi: https://doi.org/10.3390/cancers15164083.
[5] S. Koitka, L. Kroll, E. Malamutmann, et al. . “Fully automated body composition analysis in routine CT imaging using 3D semantic segmentation convolutional neural networks”. European Radiology, 31(4): 1795–1804, 2021. doi: https://doi.org/10.1007/s00330-020-07147-3.
[6] R. Hosch, S. Kattner, M. M. Berger, et al. . “Biomarkers extracted by fully automated body composition analysis from chest CT correlate with SARS-CoV-2 outcome severity”. Scientific Reports, 12(1):16411, 2022. doi: https://doi.org/10.1038/s41598-022-20419- w.
[7] J. Keyl, R. Hosch, A. Berger, et al. . “Deep learning-based assessment of body composition and liver tumour burden for survival modelling in advanced colorectal cancer”. Journal of Cachexia, Sarcopenia and Muscle, 14(1):545–552, 2023. doi: https://doi.org/10.1002/jcsm.13158.
[8] ClinicalTrials.gov. “French Clinical Database of Melanoma Patients (RIC-Mel)”. https://clinicaltrials.gov/study/NCT03315468, 2012.

Following a MSc in a field related to Biomedical Engineering or Applied Computer Science - Digital Health.