Anterior cruciate ligament injuries diagnosis using slice-aligning and multi-level feature aggregation

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Orthopedic Diagnostics

Background:

• Anterior cruciate ligament (ACL) injuries are prevalent in athletes, necessitating accurate diagnosis for effective recovery.
• Magnetic resonance imaging (MRI) is the gold standard for ACL injury diagnosis, but manual interpretation is labor-intensive and prone to errors due to slice noise.

Purpose of the Study:

• To develop an automated framework, MLFANet-SA, for detecting ACL injuries from MRI scans.
• To exclude irrelevant slices and focus on diagnostically significant regions without requiring region-of-interest or segmentation labels.

Main Methods:

• The MLFANet-SA framework incorporates a slice-aligning (SA) module utilizing local context perceptron (LCP) to identify boundary slices and unify diagnostic regions.
• A multi-level feature aggregation (MLFA) module captures spatial and cross-slice lesion patterns through channel-wise Top-K pooling and cross-slice fusion.

Main Results:

• MLFANet-SA achieved superior performance on the MRNet dataset with an Area Under the Curve (AUC) of 0.981 and accuracy of 0.949.
• On the GDAPF dataset, MLFANet-SA demonstrated strong results with an AUC of 0.975 and accuracy of 0.946.
• The model exhibited high sensitivity, specificity, precision, and MCC scores on both datasets, outperforming existing methods.

Conclusions:

• MLFANet-SA effectively combines slice selection and feature aggregation for improved localization and classification of ACL injuries.
• The framework achieves state-of-the-art performance on public and private datasets, demonstrating its diagnostic utility.
• By reducing manual labeling requirements while maintaining high accuracy, MLFANet-SA shows significant potential for aiding radiologists in ACL injury diagnosis.