Artificial Intelligence to Minimize Intraoperative Complications: Using Segmentation for Surgical Anatomy Recognition

Abstract

Lung cancer is the leading cause of cancer deaths worldwide. There is a need to develop tools to minimize intraoperative complications such as significant hemorrhage of the pulmonary artery as a result of human error. Surgeons play a role and are required to contribute to understanding newer technologies entering the field for education and improving patient outcomes. Computer vision is a viable field of deep learning that has made vast improvements in other fields of medicine like radiology diagnoses. Taking advantage of deep learning algorithms and the growing visual data collected from thoracic procedures, an object segmentation model is possible. With limited computational resources and time, experiments were conducted with the hope that results could be extrapolated for future uses such as transfer learning in the thoracic cavity. The resulting model with modifications to the data and the parameters meets all reasonable specification thresholds for educational uses. For clinical applicability and implementation, more rigorous methods of data collection and threshold must be explored. This work will help progress computer vision's entrance into more surgical fields.