Automated Bone Age Determination (ABAD) Using Wrist MRI

Honors, Awards, Invited lectures, Presentations, and Publication about this Project

By: Hamed Yousefi and Dr.Mansoor Fatehi:

This is also one of the projects by which FIFA medical assessment and research center is going to develop for determination of football player ages:

• The research has been awarded as the best paper in Iranian Congress of Biomedical Engineering. 
• Achieving the first place as the best research in 4th medical informatics conference, Feb 2013, 

The papers has been reported in international meetings including:

• CARS (Computer Assisted Radiology and Surgery), 
• European Congress of Radiology,
• Annual Meeting of European Society of Medical Imaging Informatics, 
• Turkish Congress of Radiology, 
• Future of Football Medicine (FC Barcelona), 
• Russian Congress of Radiology,
• Radiological Congress of Pakistan,

At national level it has been presented in multiple medical and engineering congresses including:

• Iranian Congress of Radiology, 
• Iranian Imaging Informatics Conference
• Iranian Congress of Biomedical Engineering, 
• Iranian Congress of Forensic Medicine
• Iranian Sports Medicine Congress, etc. 

The results of research has been also reported in private invited lectures in the following institutions: 

• Technopole (Sierre_Switzerland)
• Harvard Medical School (Boston_USA)
• Johns Hopkins Technology Innovation Center (Baltimore_USA)
• University of Southern California: IPILAB  Image Processing and Informatics Lab (Los Angeles_USA)
• Mainz University Hospital (Mainz_Germany)
• Medical University of Vienna – AKH (Vienna_Austria)  

Automated Bone Age Determination (ABAD)

Introduction

• In football there are established age-related tournaments for males and females, to guarantee equal chances within the game for all the different age groups. Over the years, the tournaments have gained momentum and popularity, particularly the under-17 competitions. Unfortunately, there has been some suspicion that the biological age of the participating players might be older than the documented age as stated by the passport or birth certificates used to determine the eligibility of the individual. This situation is aggravated by the fact that in some Asian and African countries registration at birth is not compulsory. Thus, reliable methods for proper age estimation are required. Discrepancies in age lead to unequal chances and are against the spirit of the game and, of course, “fair play”. On the other hand, different countries’ federations are very concerned about the actual age orders.
  There is cartilage-like Growth plates in wrist hand bones which ossify as people get older. One of these growth plates is located in Radius bone. The ossification of this plate is the principal criterion for legal age determination. Accordingly, through MR imaging of hand wrist, we can analyze this growth plate.
  In order to avoid unnecessary x-ray exposure in healthy young athletes, MRI has replaced conventional wrist and hand x-ray in this particular group. Besides, 3D MRI capability also provides a new way to unveil novel maturity indicators, such as growth plate volumes. With a more sophisticated technologies such as full automation of segmentation, MR volumetric of the bones would be a time-saving, robust, and reliable skeletal assessment. The individual components of our framework are outlined as shown in figure 1.

Dataset

• Dataset consists of about 1000 T1-weighted cases of hand wrist in coronal view. Based on standard protocol of FIFA, it’s usual to acquire this kind of images in 9 coronal slices. It would be better to include at least 150 cases for any 6 grades of FIFA. Also the report of expert radiologist which contains the grade of all athletics is needed.
  Since it has been researched that there are some differences between male and female ossification patterns, the available female dataset can signify the whole research study and our group would be able to study and compare the results of these cases.
  Our software is able to save all data in different formats and also release the results that include the grade of FIFA and 3D format of epiphyseal plates and its measurements related to age as PDF format.

Data Analysis

Fully Automated Procedure

• In this study, the Radius growth plate would be fully automatically segmented with high accuracy. Our method does not need the human intervention or seed points. Although segmentation techniques are developing rapidly, many recent applications are still based on available manual or semi-manual segmentation tools. Manually labeling is time consuming and inaccurate even for 2D shapes. Therefore, such a manual approach is impractical for 3D shapes.

Epiphyseal Plate Segmentation

• Advanced medical imaging techniques require high performance segmentation algorithms. Extracting the structures of interest accurately is one of the main challenges in medical imaging segmentation. The capabilities of our software would enable us to segment epiphyseal plate of radius bone fully automatically. Segmentation and 3D shape modeling of the epiphyseal plate have already been reported by our group. Dealing with the epiphyseal plate as a discrete 3D object and extracting quantitative features of this object has been considered a promising approach to understand bone age determination using MRI. A framework was designed that requires a set of MR images including the coronal slices of hand wrist information with 3mm resolution which contains 9 slices as input, and segmentation of the growth plate of Radius bone as 3D voxel points as the output of this step. The accuracy of correspondences is important, because of the sensitivity of shape parameterizations, which may lead to difficulties in defining shape constraints.

3D Visualization

• For one of the purposes of this research, visualization focuses primarily on the representation of volumetric data. The primary goals of volume visualization are to represent data within an entire volume in a manner that is both accurate and displays data as simple and intuitive way. For that reason, like any automatic advanced medical imaging techniques, high-performance segmentation algorithms are needed.

Feature Extraction

• Once a dataset has been successfully segmented, quantification of the tissue of interest is generally not a complex task. The extracted features were correlated with FIFA grade of each player interpreted by a member of AFC panel of radiologists.  The thinner the thickness of growth plate, the older the bone, and also the smaller the volume. This criterion works to determine the grade of FIFA. On the other hand, this criterion can be evaluated in any data set. The ossification pattern with aging is completely correlated with our results.

Classification Based on Grading of FIFA

• Football players are categorized into different age orders according to their ages. In many Asian and African countries, there is no access to a reliable system to determine the actual birth date, and this date can be manipulated easily. FIFA has adopted a grading system based on closure of epiphyseal plates in distal radius. According to the thickness and light intensity of the Radius bone’s growth plates, FIFA’s medical team categorizes the players’ age into one of six age classes. The purpose of this project was to explore correlations between quantitative features extracted facilitate development of a fully automated CAD system for MRI bone age determination. The SORENA software is capable to classify athletics ages based on definite FIFA grading system. But, to improve the software we need more diversion data from different grades. In the following, some achievements can reach from the available dataset with our software capabilities have been described briefly.

project goals and objectives

Improvement in the Grading System

• FIFA grading system bears some limitations in differentiating eligible vs non-eligible players. It is very probable to mark a player as grade VI while he/she has some remnants of epiphyseal plate not visualised in current protocol. Also, for U16 and U17 matches, there is no clinical guideline to decide on eligibility using FIFA grades. So, it will be expected to achieve the exact pattern of correlation between chronological age and bone age with more sample data analysis using our software.

3D Feature Analysis

• Another purpose of this project is providing a new 3D quantities features from visualization of growth plates in the Radius and Ulna bones to improve the legal age diagnosing. In proposed method, 3D visualization and feature extraction from the distal radial growth plate will provide new criteria for player’s age estimation. Combined use of separate quantitative features result in high accuracy and reproducibility.

Building 3D Atlas of Growth Plate 

• The basic idea of a shape model is to represent a class of shapes by combinations of example shapes. As with all statistical methods, the more examples are used, the better the model can represent the class of objects it models. Statistical models are able to capture the variability of shapes in a dataset.
  Probabilistic atlases derived from the information of a plentiful number of subjects represent structural information of the subject’s population.  The produced atlas can be used as a primary knowledge of image processing algorithms or be used to detect special irregularities of a group of subjects. This model should be flexible enough to include new data, and using more data concludes to an easier determination of variations. Probabilistic atlases contain information about variation amounts in each voxel, but don’t tell us about the type of the variations. Statistical atlases give us this information about variation types. The main aim of this part is to develop a morphometric and anatomically accurate atlas (statistical shape model) of growth plate inside of Radius bone for better observation of the ossification pattern with age. A procedure for building atlas included fully automatic segmentation of the growth plates, extraction and identification of common features, averaging feature position, obtaining mean geometry, mathematical shape description and variability analysis

Ossification Pattern Study

• Skeletal age is determined from the development stage of bones. These approaches estimate development stages from the fusion/maturation of specific bones. The pattern of ossification in the hand and wrist bones is in a fairly predictable manner and age-specific until the end of adolescence when the elongation of bone is complete. Thus, the standards of bone age have been derived by comparing the level of maturation of hand and wrist bones with normal age levels. We are going to analyze the pattern of ossification of epiphyseal of radius with age. Having these results, we can sense 3D shape variations of epiphyseal plate. For that reason, the reasonable cases with different ages are needed.