BME.09 – Bleeding Point Detection System For Visualization During Endoscopic Spine Surgery

• Omar Al Mhtachem
• Ashley Anderson
• Hisham Elseweifi
• Amirali Monshizadeh
• Manav Shah
• Nameera Umar

Endoscopic spine surgery (ESS) has emerged as an ultra-minimally invasive surgical technique for various spinal procedures, offering advantages such as shorter recovery times and reduced tissue damage. However, ESS faces challenges, particularly in bleeding control and consequential limited visualization. This project focuses on the development of an innovative software-based visualization model designed to enhance real-time bleeding point identification in the context of ESS. The need statement is refined to acknowledge the overarching objective of sustaining a clear visual field in the presence of a bleed and reducing the surgical procedure time, recognizing that addressing bleeding is not solely about locating the point of bleeding but also about optimizing overall visibility and efficiency. The design objective has evolved to emphasize the creation of a comprehensive visualization tool that actively contributes to the surgeon’s capabilities during ESS. The design criteria have been established with meticulous consideration of factors such as minimal disturbance, compatibility with existing endoscopes, automation, minimal latency, and high sensitivity and specificity to bleeds has been undertaken. These criteria are essential to ensure seamless integration into the surgeon’s workflow and minimal interference during the procedure. The proposed solution leverages a sophisticated deep learning (DL) model utilizing semantic segmentation, a cutting-edge technique associating each pixel with a specific class, facilitating the precise identification of bleeding points. Initial results showcase the effectiveness of the DL model in identifying bleeds, paving the way for a significant advancement in ESS technology. The use of neural network models to identify bleeding endoscopic surgery shows promising results. Our verification method indicated that the model can correctly identify the bleeding point in vast majority of test frames. The model labeled the bleeding point (n=20) within 5mm from the center of the true bleeding point with statistical significance (p = 9.74E-09 < 0.05, from t = -9.21 < 2.086). The average error across all frames was 1.9423 mm. The holistic approach of this software-based visualization model, along with its emphasis on real-time bleeding point identification, positions it uniquely within the industry. As the adoption of endoscopic spine surgery continues to grow, this innovative solution has the potential to lower barriers, increase adoption rates, and contribute to the evolving realm of spinal surgical technology in the United States.