Arthroscout is a tool classification software that utilizes image processing and machine learning methods to detect and classify the types of tools used in the arthroscopic surgery. An arthroscopic surgery, also known as "the keyhole surgery", is an operation where surgeons examine and treat a joint by inserting an arthroscope, a pencil-sized instrument equipped with a miniature camera. On average, each orthopedic surgeon performs about 300 to 500 arthroscopies each year. For each arthroscopy, 3 to 8 videos are recorded for the purpose of documentation and diagnosis, resulting in 900 to 4,000 videos recorded per year per surgeon. Worldwide, over 2 million arthroscopy procedures are performed annually. 

Browsing and annotating dozens of videos is a time consuming task for surgeons. Our goal is to lessen the workload of the surgeons, and allow them to focus more on effectively treating patients. We believe that the high accuracy, tool classification software is an important step for video summarization and automatic annotation. Since each tool has unique functionalities, identifying tools within videos can link relevant events to help summarize videos. 

The goal of this project is to create a software that can identify surgical tools in a real-time video display to perform the task of video summarization/annotation. The algorithm is implemented by constructing and training a convolutional neural network using Python’s Tensorflow library to identify the various tools.