Date of Award

6-2025

Degree Name

Doctor of Philosophy

Department

Electrical and Computer Engineering

First Advisor

Janos L. Grantner, Ph.D.

Second Advisor

Ikhlas Abdel-Qader, Ph.D., P.E.

Third Advisor

Saad A. Shebrain, MD

Abstract

Manual evaluation of suturing skills during laparoscopic training is often subjective and labor-intensive, resulting in the lack of scalable and consistent feedback for trainees. This study proposes an automated framework that not only significantly reduces the need for in-person assessment by experts but also ensures scalability, thereby addressing the objectivity and cost-effectiveness limitations. While low-cost laparoscopic box trainers have become increasingly popular for residency training, performance assessment still depends on expert supervision. The proposed system aims to alleviate these limitations.

This study introduces a novel automated framework incorporating an optimized DeepSORT algorithm for classifying, localizing, and tracking surgical tools using video data. This framework is designed to address the limitations of in-person assessments and provide scalable and consistent feedback for trainees. A fuzzy inference system (FIS) was developed to allow for suturing performance assessment using surgical suturing key metrics such as motion quality, idle time, completion time, tissue status, and upward force. Twelve videos of various suturing tasks were recorded and annotated using the Intelligent Box Trainer System (IBTS) in the Intelligent/Fuzzy Controllers Laboratory, creating a dataset for surgical tool tracking and performance assessment. The proposed framework integrates object detection and tracking to analyze surgical tool movement and reconstruct accurate needle paths in 3D. The YOLO architecture detector is used for real-time object detection followed by two algorithms, optimized DeepSORT and centroid tracking, to track the movements of the tools. To generate highly accurate 3D needle paths, the system uses the inverse projection transformation to map 2D video data into spatial coordinates. As a result of the high accuracy in identifying needle and tissue positions (80.1% for YOLOR and 89.6% for DeepSORT), force sensitivity and idle time were obtained.

The fuzzy inference system (FIS) utilizes parallel and cascade architectures to integrate deep learning results and measurement metrics for a complete, automated framework. Input parameters such as knots tied, tissue sealing success, force applied, smoothness of the instrument, and operating times are all processed simultaneously by the FIS. A centroid defuzzification method is also implemented to convert the fuzzy outputs into precise numerical scores, facilitating performance interpretation. This integrated system measures surgical trainee proficiency accurately and objectively, which improves outcomes and reduces manual evaluations.

The FIS system integrated with the automated framework improves overall assessment accuracy by 18%, enhancing the system's ability to provide consistent, and objective assessments. This study is a significant step in developing an affordable, computer-aided, and functional laparoscopic training and assessment system based on a laparoscopic training box equipped with webcams. Future work will focus on expanding the dataset and enhancing real-time feedback to support adaptive learning.

Access Setting

Dissertation-Open Access

Download

Included in

Biomedical Commons

Share

COinS