Evaluation of Transfer Learning Efficacy for Surgical Suture Quality Classification on Limited Datasets

Area of Science:

• Medical Imaging and Artificial Intelligence
• Surgical Skill Assessment

Background:

• Objective assessment of surgical skill is crucial but often relies on subjective evaluations.
• Limited availability of medical data presents a significant challenge for developing automated diagnostic tools.

Purpose of the Study:

• To evaluate the effectiveness of transfer learning using pre-trained Convolutional Neural Networks (CNNs) for automated binary classification of surgical suture quality.
• To assess classification performance across three distinct suture types: interrupted open vascular sutures (IOVS), continuous over-and-over open sutures (COOS), and interrupted laparoscopic sutures (ILS).

Main Methods:

• Eight state-of-the-art CNN architectures were trained and validated on small datasets (100-190 images per type) using 5-fold cross-validation.
• Performance metrics included F1-score, Area Under the Receiver Operating Characteristic Curve (AUC-ROC), and a custom weighted stability-aware score (Score_adj).
• GradCAM visualizations were employed to interpret model focus on clinically relevant features.

Main Results:

• Transfer learning demonstrated robust classification performance despite data scarcity, achieving F1 scores > 0.90 for IOVS/ILS and 0.79 for COOS.
• ResNet50V2, DenseNet121, and Xception exhibited higher stability based on Score_adj.
• ResNet50V2 achieved the highest AUC-ROC (0.959 ± 0.008) for IOVS internal view classification.

Conclusions:

• Transfer learning is a powerful and accurate approach for developing objective, automated surgical skill assessment tools, even in resource-constrained settings.
• The study validates the use of CNNs for analyzing suture quality, reducing the need for subjective expert judgment.
• Visualizations confirmed that models focus on clinically significant features, enhancing the reliability of automated assessment.