Surgical Tool Detection: A Comparative Study of Supervised and Semi-Supervised Learning Approaches

Shiva Shokouhmand, Md Motiur Rahman, Smriti Bhatt, Suranjan Panigrahi, Miad Faezipour

Abstract: Surgical tool detection and classification is of great importance in enhancing the efficacy, safety of surgical procedures, and providing advanced surgical assistance, especially applicable to robotic/automatic surgery. Current methodologies predominantly rely on supervised techniques necessitating extensive annotated datasets. However, the availability of labeled datasets remains a persistent challenge. In this research, the objective is to detect and classify surgical instruments while evaluating the performance of supervised machine learning and deep learning models against a semi-supervised approach in image classification. Also, we investigate and contrast the significance of preprocessing, transfer learning, and semi-supervised learning techniques. Our findings demonstrate that while thorough preprocessing mechanisms can significantly impact the performance of machine learning models, employing transfer learning with deep learning models yields superior results. Additionally, training data on semi-labeled datasets exhibits comparable, or in some cases better performance than supervised deep learning methods, particularly in scenarios where labeled data is scarce.

External IDs:doi:10.1007/978-3-031-85908-3_29