Safer Surgeries Using AI: Enhancing Surgical Accuracy with Machine Learning Algorithms for Instrument Counting

This study explores the integration of Convolutional Neural Networks (CNNs) into surgical procedures to address the critical issue of retained surgical instruments (RSIs), enhancing patient safety. RSIs are when a foreign object is unintentionally left within a patient after surgery or other invasive procedures. This paper discusses challenges associated with manual counting methods and the potential of CNNs in automating surgical tool recognition. It employs an experimental design to investigate the impact of model hyperparameters on machine learning performance, utilizing the "Labeled Surgical Tools and Images" dataset for training. Preprocessing techniques, data augmentation, and Dropout regularization enhanced model robustness. Results from multiple training trials demonstrate the efficacy of the CNN-based model in accurately identifying and classifying surgical instruments, even in limited data scenarios. This research identifies overfitting as a challenge and addresses it through model adjustments and regularization techniques; it also highlights the findings’ implications for improving surgical instrument count accuracy and enhancing patient safety. This study concludes by emphasizing the transformative potential of CNNs in surgical practice and the importance of ongoing research further to advance machine learning technologies in real-world surgical settings.