Combined analysis of image data from head and neck cancer

Head and neck cancer is a global health problem with an average of 890,000 new cases and 450,000 deaths per year. Despite ongoing research into diagnosis and treatment options, the overall five-year survival rate has not significantly increased in the last decades. One major challenge lies in the heterogeneity and complexity of head and neck cancers, which comprise a diverse group of tumors that arise from different cell types and anatomical locations in the head and neck region. In case of surgical treatment, up to now the main challenge is a reliable intraoperative assessment and the complete resection of the tumor. Currently used intraoperative in vivo imaging devices are lacking the ability to clearly define the tumor margins. In this work I analyzed head and neck squamous cell carcinomas (HNSCC) and salivary gland carcinomas (SGC) and combined novel spectral and molecular imaging methods with deep learning based data analysis to overcome clinical challenges in treatment and diagnosis of HNSCC and SGC. The two imaging methods were matrix assisted laser desorption/ionization imaging (MALDI) and hyperspectral imaging (HSI). MALDI imaging aims to provide a label- free, spatially resolved measurement of the abundance of biomolecules by using the difference in their mass-to-charge ratio. This leads to a mass spectrum for each location of a given sample. HSI on the other hand measures the interaction of a sample with a broad light spectrum ranging from the visible to the infrared. This thesis is based on three studies. In the first two studies, I was able to show that complex medical problems in head and neck cancer diagnosis and treatment can be addressed by combining novel imaging techniques with deep learning based data analysis. In the third one, I helped to show, that there is no tumor specific uptake of ICG by combining and analyzing fluorescence microscopy and classical histopathological data.