Automated knee ligament injuries classification method based on exemplar pyramid local binary pattern feature extraction and hybrid iterative feature selection

Background: Knee ligament injuries have been widely seen in orthopedics and traumatology clinics worldwide. A correct diagnosis is required for treating knee ligament injuries diseases as with other diseases. Magnetic resonance images (MRI) have been often used for the diagnosis of knee ligament injuries. Problem definition: Automated disease detection methods must be used in clinics to save more time and help medical doctors with diagnosis. This research aims to present an intelligent assistant system to detect knee ligament injuries automatically. Method: This research presents a new hand-crafted feature generation, and this feature generation model is the exemplar pyramid local binary pattern (LBP) technique. A hybrid feature selector is applied to the generated features for selecting the most valuable/informative features. This feature selector uses ReliefF and Iterative Neighborhood Component Analysis together. The prime objectives of this feature selector are both to select the optimal number of features and using effectiveness both ReliefF and NCA. Two shallow classifiers are used to denote strength both feature generator and used hybrid feature selector. The presented model is tested on three MRI datasets about knee ligament injuries. Results: The proposed exemplar pyramid LBP and RFINCA based automated classification method reached 99.32%, 99.56%, and 100.0% classification accuracies for the collected three datasets respectively using the KNN classifier. Conclusions: These results demonstrated the general and high success of this method. The obtained results were also shown that an intelligent health assistant for knee injuries could be developed by using the proposed exemplar pyramid LBP method.

Automated summary

The study proposed an intelligent assistant system for automated detection of knee ligament injuries, achieving high classification accuracy on MRI datasets.