ORALHYPE 2.0 - Oral Health with Hyperspectral Imaging and Computer Vision. Approach to establish non-invasive hyperspectral sensoring in the surgical therapy of oral squamous cell carcinoma by the use of computer vision

• Applicants: Professor Dr.-Ing. Thomas Klauer; Dr. Daniel Thiem
• Subject Area: Dentistry, Oral Surgery ; Medical Informatics and Medical Bioinformatics
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 516210826

The aim of this project is the establishment of hyperspectral sensoring as a radiation-free, non-invasive biophotonic methodology to assess the dignity of (oral) mucosa lesions. Primary focus is on the diagnostic assessment of the oral squamous cell carcinoma pericenter for perspective evaluation and determination of the extension of its surgical safety margin during tumor resection using computer vision as a tool for automated image analysis. This, as a scientific discipline at the border between computer science and engineering, uses techniques of artificial intelligence (artificial neural networks and deep learning) for hyperspectral image dataset processing, enabling IT-based automated tissue classification based on the specific hyperspectral tissue signature. Besides the creation of a hyperspectral database of oral ground tissue types (muscle, fat, mucosa) based on ex vivo (already investigated and published) and in vivo hyperspectral 3D image information, main milestone is the creation and IT-supported processing of ex vivo and in vivo image datasets of oral squamous cell carcinomas and their peritumoral mucosal margins using different spectrometer variants (hyperspectral endoscopy unit and hyperspectral table top camera). In addition to the automated classification of ground tissue types, this project aims to provide a computer-aided classification of pathological tissues into severity grades on the basis of their tissue structure and on the basis of their typical hypersperctral signature. We plan to develop a non-invasive measuring technique for the treatment of oral squamous cell carcinoma by evaluating hyperspectral image data using computer vision, which will solve a major problem of the current state of the art. In order to increase the efficiency of IT-supported real-time evaluation of acquired image data sets, suitable image processing methods have been discussed in science and practice for quite some time with limited training data availability. However, large-scale application currently still fails due to the lack of suitable concepts for transferability of these algorithms and methods to the field of hyperspectral technology. The application of supervised and unsupervised learning methods therefore has considerable potential if successful.

• DFG Programme: Research Grants
• Co-Investigators: Professor Dr. Peer Kämmerer; Dr. Stefanie Zimmer