Predicting Postoperative Re-Tear of Arthroscopic Rotator Cuff Repair Using Artificial Intelligence on Imbalanced Data

Retears after rotator cuff surgery are a common complication. Accurate prediction of retear is essential to minimise the risk of retear. Most of the existing big methods for assisted diagnosis are tested on balanced datasets, whereas in the medical field there is data imbalance due to the rare natur...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhibin Zhang, Zhewei Zhang, Zhaoxiang Peng, Yihong Dong
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10870224/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Retears after rotator cuff surgery are a common complication. Accurate prediction of retear is essential to minimise the risk of retear. Most of the existing big methods for assisted diagnosis are tested on balanced datasets, whereas in the medical field there is data imbalance due to the rare nature of a particular disease or case. Therefore, this paper proposes a weight-based logits cross-entropy loss function. The model in this paper is mainly divided into two parts. This paper employs a pairwise associative encoder (PAE) to construct edges, calculating similarity scores between different nodes to effectively handle the complementary relationship between phenotypic and non-phenotypic data. The other part is the weight network, which calculates the weights of different nodes of different classes to assign weights to the objective function through the GCN model. The main body of the final objective function is the weighted cross-entropy loss function, while the logits of the model output are adjusted and a regularisation term is added to the objective function in order to deal with the discrete points in the data. The classification performance of the proposed method is tested on the dataset of whether rotator cuff is re-torn after surgery, and the results show that our method achieves state-of-the-art performance in a five-fold cross-validation setting.The code of this work is available at <uri>https://github.com/nishuihan111/WL-GCN.git</uri>.
ISSN:2169-3536