The Healthcare domain has seen a wider use of classification models to predict whether the individual would have heart disease (yes or no), diabetes (diabetic or non-diabetic), or tumour (benign or malignant). There has been one question arising a lot whether we could predict any sort of injury or not through any sort of machine learning model.
How much it is pain for a spectator to look at his or her favorite player coming out of the bout due to injury?
The ACL is a band of thick connective tissue that flows from the tibia to the femur. As it withstands anterior tibial translation and rotational loads, it is considered a crucial structure in the knee joint. The prevalence of an anterior cruciate ligament is significantly high in both contact and non-contact sports. Most commonly, they occur in those who play sports involving pivoting, jumping, and deceleration (e.g. football, basketball, handball, and athletics). They can manifest themselves in different ways from mild (like small tears) to severe (when the ligament is completely torn).
When so ever there is the incidence of injury report; there is a broad impact in the team if it is a team sport or individual in case of individual reports. We take into consideration economic factors associated with many sorts of injuries but how about psychological impact; impact in a team due to absence of key player due to injury that could later cost the match and henceforth medal or winning the cup.
How to predict ACL injury?
As stated above Anterior Cruciate Ligament injury occur during pivoting, jumping, and deceleration. Whenever there is a high loading in knee injury there would be a likelihood of Anterior Cruciate Ligament injury. Therefore; we could build a model that could provide a binary answer whether there is a high knee loading or not. If there is a high knee loading there are chances of getting ACL injuries in that individual athlete.
In research, it has been proven that if there is high knee loading that could lead to greater translation and causing injury.
A feature selection has to be the most vital step during the building of the model to predict ACL injury as several factors may account for it. A brief overview of features (factors or parameters):
- Gender: Females are three times more prone to have an ACL injury than males due to smaller size and different shapes of the intercondylar notch, biomechanical variation (Q-angle and wider pelvis), and greater ligament laxity due to the influence of hormones.
- Knee Valgus and Flexion Angle: The standard ACL injury occurs with the externally rotated knee and bending at 10-30 ° as the knee is put in a valgus position when the athlete takes off from the planted foot and rotates internally to change direction abruptly. It has to be checked during drop jump with the implication of Video Analytics.
- Muscle Strength: An isokinetic dynamometer and electromyography analysis of quadriceps (front thigh muscles), hamstrings (back thigh muscle), and gastrocnemius (below knee back muscles) would provide an objective measure of strength. In Isokinetic Dynamometer the strength has to be checked in the concentric phase (a type of muscle work) at 300 degrees per seconds’ angular velocity. A surface electrode has to be placed in the motor point of the gastrocnemius while doing concentric and eccentric work of muscle. EMG Analysis has to be done during drop jump.
- Anthropometric parameters such as tibial length and weight also have to be taken into account.
There are few other parameters like knee laxity during the Lachman test (orthopedic special test) by using knee laxometer, Q-angle, Shoe Surface Interface, and Type of Sports could also be taken into consideration.
Model Building and Selection
A classification model will be built such as Logistic Regression, Naïve Bayes, Support Vector Machine, Decision Tree, Random Forrest, Stochastic Gradient Descent, and K- Nearest Neighbors to predict the label data as high and low knee loading (a binary classification). A Grid Search CV could also be built to do hyperparameter tuning to get better accuracy, sensitivity, and specificity.
During a model evaluation, we have to take into consideration of sensitivity (true positive rate) and specificity (true negative rate) through confusion matrix and classification report, we have to consider a model that has high sensitivity score (a test with 100% sensitivity will recognize all-athlete with the high knee loading by testing positive).
The significance of building an ACL prediction model will help the team and individual to predict the ACL injury. If we found the outcome as high knee loading so then we could recommend the particular athlete to go for strength and conditioning training, henceforth, it would help economically and mentally also.