Automatic Bone Age Determination Using Wrist MRI Based on FIFA Grading System for Athletes: Deep Learning Approach

Purpose: The Fédération Internationale de Football Association (FIFA) has developed a grading system consisting of levels I to VI that can be used in teenage athletes. The grading system is currently used as the standard method to determine bone age in football players. All national and club matches are obliged to follow screening procedures strictly (e.g., anti-doping procedures). This study evaluated the performance of an automatic system based on deep learning that provides a FIFA grade upon receiving Digital Imaging and Communications in Medicine (DICOM) images of a magnetic resonance imaging (MRI) study to facilitate and speed up the determination of bone age.

Methods and Materials: The FIFA bone age determination system consists of six grades starting from a totally unfused epiphyseal plate (grade I) to a completely fused plate (grade VI). Variable progressive degrees of fusion are considered the basis for grades II to V. The protocol includes 9 slices in the coronal plane with a 3-mm gap between the slices. The recommended MR sequence is T1. Because the middle image in the nine-picture data set is considered the most informative slice containing the largest image of the distal radius, the study was done using this single slice as the basic source of grading, and then another volumetric set of slices 4, 5, and 6 were also used as the second group. A convolutional neural network was designed in four convolutional layers including pooling, rectified linear unit (ReLU), and fully connected layers. Fifty-five teenage football players of the national U17 team were examined using a 1.5-T Siemens Avanto machine. The studies were interpreted by a musculoskeletal radiologist member of the AFC panel of radiologists aware of the FIFA scoring and grading system. Thirty-six cases were used for training and 19 cases for testing of the CNN. To increase the number of training images, augmentation was performed by rotation and moving the original images. Overall, 613 images were obtained for training and 267 for testing.

Results: Images were introduced to the neural network resulting in sequential layers of meaningful output. The final outcome of the network as the FIFA grade of the case was compared with the interpretation of the radiologist (Table 1). The findings indicate a high accuracy of the single-slice data set; the accuracy approached 100% when the volumetric three-slice sets were used.

Conclusion: The findings of this research indicate a convolutional neural network can be used for automated bone age determination and FIFA grading of wrist MRI with reasonably high accuracy.

Table 1

Accuracy		Group
Volumetric three middle slices, %	Single middle slice, %
99.62	97.75	Overall
100	90	Grade II
98.91	98.91	Grade III
100	97.91	Grade IV
100	97.43	Grade V
100	100	Grade VI