Early detection of head and neck tumors is crucial for patient survival. Often, diagnoses are made based on endoscopic examination of the larynx followed by biopsy and histological analysis, leading to a high interobserver variability due to subjective assessment. In this regard, early non-invasive diagnostics independent of the clinician would be a valuable tool. A recent study has shown that hyperspectral imaging (HSI) can be used for non-invasive detection of head and neck tumors, as precancerous or cancerous lesions show specific spectral signatures that distinguish them from healthy tissue. However, HSI data processing is challenging due to high spectral variations, various image interferences, and the high dimensionality of the data. Therefore, performance of automatic HSI analysis has been limited and so far, mostly ex-vivo studies have been presented with deep learning. In this work, we analyze deep learning techniques for in-vivo hyperspectral laryngeal cancer detection. For this purpose we design and evaluate convolutional neural networks (CNNs) with 2D spatial or 3D spatio-spectral convolutions combined with a state-of-the-art Densenet architecture. For evaluation, we use an in-vivo data set with HSI of the oral cavity or oropharynx. Overall, we present multiple deep learning techniques for in-vivo laryngeal cancer detection based on HSI and we show that jointly learning from the spatial and spectral domain improves classification accuracy notably. Our 3D spatio-spectral Densenet achieves an average accuracy of 81%.
Here we present a study where we used in vivo hyperspectral imaging (HSI) for the detection of upper aerodigestive tract (UADT) cancer. Hyperspectral datasets were recorded in 100 patients before surgery in vivo. We established an automated data interpretation pathway that can classify the tissue into healthy and tumorous using, different deep learning techniques. Our method is based on convolutional neural networks (CNNs) with 2D spatial or 3D spatio-spectral convolutions combined with a state-of-the-art Densenet architecture. Using both the spatial and spectral domain improves classification accuracy notably. Our 3D spatio-spectral Densenet classification method achieves an average accuracy of over 80%.
Tumors of the upper respiratory tract are the sixth most common tumor entity in humans. Currently a dedicated screening method enabling a direct onsite diagnosis is missing. This can lead to delayed diagnoses and worse outcomes of the patients. An optical method enabling a direct distinction between healthy tissue, dysplastic tissue and cancerous tissue would be an ideal tool for the detection of tumors of the upper respiratory tract. In this study we used fluorescence lifetime imaging (FLIM) of NADH and FAD to image the metabolic state in different tissue samples of the upper aerodigestive tract (UADT). Due to the different metabolic pathways that are active in healthy and tumor cells their metabolic states differ significantly. FLIM datasets of tissue samples from 25 patients were recorded directly after surgery ex vivo in a special tissue culture medium at 37°C on a dedicated microscope using multiphoton excitation. By calculating the fluorescence-lifetime redox ratio (FLIRR) based on the FLIM measurements, we were able to visualize the metabolic state of the cells. We found that healthy tissue, dysplastic tissue and cancerous tissue showed significant differences in the FLIRR. This study suggests that the FLIRR might be a sensitive and robust parameter for the differentiation of cancerous and pre-cancerous UADT tissue and that optical metabolic imaging could be a valuable tool for an early tumor diagnosis within this area.