Physicians and surgeons often face problems with identifying critical structures in preparation for surgeries. Brutenis Gliwa, a working student at PLANET AI, worked on an approach to support surgeons with this task by using algorithms.
Brutenis recently finished his bachelor thesis „Automatic Classification of the Bronchus Structures Using a Rule-Based Approach“ in computer science at the Department of Systems Biology and Bioinformatics (SBI, University of Rostock). There, a team of multidisciplinary scientists led by Prof. Olaf Wolkenhauer researches in the fields of clinical diagnostics, data analysis and more. Brutenis’ thesis was supervised and supported by Mariam Nassar and Gundram Leifert on the side of PLANET AI and Dr. med. Rolf Oerter, specialist for thoracic surgery at University Medicine Rostock.
As a member of the AIrway project at the SBI, Brutenis developed a pipeline that automatically classifies the bronchial structures of the human lung. Machine Learning came to use in an external step, where CT scans provided by Dr. Oerter were transformed into bronchus masks. Using these masks and the open-source 3D graphics software Blender, Brutenis developed and applied a vector-based rule set to assign colors to the bronchus (see video on the right).
As a result, physicians can use the classified models to plan surgeries more precisely. Ultimately, patients could benefit since only as much tissue as needed is removed from the lung. This applies to thoracic surgery as of now, since AIrway focuses on processing and structuring thoracic scans. Further research and development could include using the software for bronchoscopy or comparing patient data for statistical analyses. The created ground truth data could also be used for the training of an AI-based approach. On Github, you can find the AIrway source code.
The cooperation of PLANET AI and the University Medicine of Rostock is a vital part of Doctor AI, our approach for a support system for telemedicine, emergency care, and more.
Learn more about research at PLANET AI.