In In glioma surgery, maximum tumor resection and minimum complications are important. An expert surgeon will resect the glioma based on the brain structure, function and tumors position for each patient, by considering postoperative complications. Therefore, surgical process, work contents and duration for surgery vary by cases. Hence, it is difficult for young surgeons and surgical staff to understand the surgical process and to predict the next work in awake surgery for glioma. We aim to develop the system of the surgical process identification that enables young surgeons and surgical staffs to understand surgical processes of the expert surgeon in real time during awake surgery for glioma. In this study, we propose of surgical workflow modeling and identification of surgical processes methods using machine learning by information obtained from multiple medical devices in operating room. First, we created the surgical processes model which has 12 surgical processes. Seconds, we automatically extract the features of surgical processes using medical image processing and YOLO (You Only Look Once) of machine learning by pre- and intra- operative images, navigation system’s log and microscope video. Finally, surgical processes are identified using HHMM (Hierarchical Hidden Markov Model). To estimate our method, we evaluated surgical process identification accuracy and system processing time using 3 past clinical cases in awake surgery for glioma. Then, it was shown that surgical processes can be identified with high accuracy while ensuring the real time properties of a processing.
Accurate force measurement during forceps manipulation is expected to have various applications such as surgical technique analysis. The calibration method is important for achieving accurate measurement. Previously, for a force measurable forceps (FMF) with 3 degrees of freedom (DOF), a calibration method using unbiased samples has been proposed. However, the required number of samples increases exponentially as the DOF increases. Here, we proposed a semi-automated sampling system for collecting unbiased samples from a FMF with 4DOF. We conducted a collection of unbiased samples, calibration, and compared the accuracy with the standard calibration method for linear force sensors. Sampling took around 10 seconds on average (n=1535). The accuracy, evaluated by the average error[N], for the unbiased sample calibration method (radial traction: −0.0565±0.0638, axial traction: −1.51±4.32, grip: 0.780±1.00) improved by approximatively two folds compared to the standard method (radial traction: −0.112±0.132, axial traction: −8.60±9.04, grip: −1.17±1.07), with maximum measurement ranges[N]of±2.00 for the traction, and±1.60 for the grip. We conclude that these results show the efficiency and accuracy of the proposed device, when compared to the conventional standard methodology.