Age estimation in living individuals from MRI images of the knee
Research type
Research Study
Full title
Age estimation in living individuals using MRI images of the knee from a Scottish population: development of a new method for age assessment based on the distal end of the femur and development of a protocol of study based on analysis of the MRI images.
IRAS ID
215722
Contact name
Lucina Hackman
Contact email
Sponsor organisation
University of Dundee
Duration of Study in the UK
2 years, 9 months, 29 days
Research summary
Summary of Research
Age is an essential parameter to evaluate in the legal and immigration context when a person cannot provide valid documentation proving his/her chronological age and it becomes necessary to establish if he/she has reached a specific legally relevant age threshold or not. In living individuals it is possible to assess age utilising skeletal development. This is usually done through a radiograph of the left hand. There are, however, limitations in the use of radiographs, especially since they involve exposure to ionising radiation. In response to this, non-ionising imaging technologies (MRI and ultrasound) are beginning to be tested for age estimation purposes. As the research into the use of MRI scans for age estimation is just beginning, few studies exist and more research is needed to ensure that age estimation methodologies based on MR images are accurate and reliable. \nThe project will aim to create a valid age estimation method based on MRIs of the knee and to establish a protocol for practitioners to follow when utilising MRIs to assess age. \nThe first part of the research will focus on the description of the development of the distal end of the femur using Proton Density Fat Suppressed MRI images, followed by the development of a staging method that can be used for age estimation. The second part will focus on the comparison of images taken with different MRI settings and the development of a protocol of analysis.\nFor the realisation of this work, the MRIs of a max of 440 individuals between 10 and 20 years will be studied. \nThe research will last 4 years, during this time there won’t be any contact with the patients. The images that will be used are from Tayside patients who have undergone MRI examination for medical reason at Ninewells Hospital (Dundee). No identifiable personal data will be gathered, the only information that will be gathered is sex, date of birth and date of images.\nSummary of Results
The aim of the research was to describe maturation of the distal end of the femur using Magnetic resonance imaging (MRI) of the knee, relate the maturational changes to chronological age, and investigate the possibility to develop an age estimation method that can be used in a forensic context in cases when the age of an individual has to be assessed. The aims were achieved by studying the T1Weighted and Proton Density Fat Suppressed MRIs of a total of 321 individuals, 164 males and 157 females, from a Scottish population (Tayside patients), with an age between 10 and 20 years.
The research was divided into 5 different phases, each one depending on the previous one and important for the progression of the research.
The first phase of the research aimed at developing the classification system used to analyse the T1W MRIs. As result of the first phase, an 8 stages classification system was created and tested for reproducibility via inter and intra observer error. The classification system was then used to describe, record and collect information regarding the maturation of the growth plate of the distal end of the femur in each individual. The data collected for each MRI were used to create maturational maps which allowed to see exactly where maturation was happening within the growth plate for each individual.
The second phase of the research aimed at having an initial overview of maturation of the distal end of the femur across the different ages and to determine the most common stages in each 1-year age cohort, based on the use of T1W images. The MRIs were analysed separately for male and female individuals at all times, and for this phase of the research, the individuals were divided into 1-year age cohorts. The division into 1-year age cohorts was important as one of the overall aims of the research was to assess maturation at forensic relevant age thresholds such as 12, 14, 16, and 18 years of age. As a start, during the second phase of the research, all MRIs were analysed using the classification system developed in phase 1 of the research. During the analysis of the MRIs, it was observed that fusion of the growth plate would follow an apparent pattern, that different areas of the growth plate would mature at different times, and bone bridges would start appearing in the central areas of the growth plate, rather than at the borders of the growth plate. To further investigate the fusion pattern of the growth plate, the growth plate in each maturational map was divided into
defined areas that could then be compared and that could be of assistance in the description of maturation. The most common stages for each area for each age cohort were also calculated and used to obtain representative maturational maps for each age cohort. It was then investigated if there were presence of significant differences between the areas of the growth plate, and, if present, where those differences were.
MRI images that could represent the maturational stage of each age cohort, showing location and most common stages present within the growth plate were also selected. This was done so that the appearance of the maturational maps could be related to the appearance of the growth plate in the MRIs. The results showed that there was presence of significant differences between the areas of the growth plate between the central areas of the growth plate and the most peripheral areas, especially at the younger ages. The initial observations regarding the presence of a pattern of fusion of the growth plate, starting at the centre and progressing centrifugally were also confirmed using the maturational maps representative of each age cohort.
The aim of the third phase of the research was to complete a detailed description of maturation of the distal end of the femur, identify the pattern of fusion of the growth plate, provide age ranges for each of the 8 stages, and compare maturation between male and female individuals. This phase allowed to set the basis also for the final stage of the research using T1W MRIs, phase 4, phase dedicated to the development of the age prediction method. During phase 3, the division by age categories was dropped to allow comparison across all ages. The different areas of the growth plate, as defined in phase 2 of the research, were kept based on the results obtained in phase 2 which showed presence of significant differences between the areas at different ages. The areas of the growth plate were studied and compared. The results confirmed what initially observed in phase two. The first bone bridges would appear at the weight bearing areas of the growth plate, where it is necessary to ensure stability and contrast the shearing forces the bone undergoes normally. Fusion then progresses centrifugally, with the borders of the growth plate fusing last. Maturation and appearance of the bone bridges, was then related to chronological age, and age ranges for each stage for each area of the growth plate were obtained. This information was used to see if the appearance of certain stages, along with their location was useful to determine minimum age of an individual or determine if a relevant age threshold was reached. A table summarising the stages useful to assess the ages of 12, 14, 16 and 18 years was created so that it can be of assistance when assessing the age of the individuals undergoing the age assessment. Furthermore, significant differences between male and female individuals regarding the time of appearance, presence and distribution of the different stages across time were observed. It was also estimated that female individuals were ahead in maturation compared to male individuals of approximately 13 months.
The fourth phase of the research aimed at developing age prediction models that could be used to predict the age of individuals who needed their age to be assessed. The analyses were based on the results obtained during phases 2 and 3 of the research. Furthermore, it was tested if by using location of each stage the age prediction would improve, and if similar predicted ages would be obtained if only selected areas of the growth plate were used instead of the whole growth plate. By using appropriate algorithms, it was also established what were the stages each model was relying upon the most to predict age. The results were then linked back to the results obtained in phase 3, and to the age ranges these stages were present and to their distribution across the ages. The aim was to identify specific stages that could be key to determine the age of an individual or key to specific age thresholds. As result, six ageprediction models for the male and 6 prediction models for the female individuals were developed using machine learning techniques. The age prediction models were then tested on 40 MRIs (20 MRIs for the male and 20 MRIs for the female individuals) as proof of concept, to confirm that the model work on new MRIs and that they are suitable to predict age. From the analysis it was observed that presence of stage 7 within the growth plate, corresponding to complete fusion of the growth plate, was the stage each model was relying upon the most for predicting age. No improvement in the age assessment was observed if location of the stages was considered when predicting the age, compared to when location was not used. The best age prediction was obtained when considering the whole growth plate, rather than when using selected areas.
The final phase of the research involved the analysis of the PDFS MRIs and their comparison with T1W MRIs to see if by using 2 MRI imaging modalities acquired from the same individual, the age estimation prediction would have improved. To start, a 3 stages classification system for the analysis of PDFS MRIs was developed and tested for intra observer error. The PDFS MRIs were then analysed using the newly developed 3 stages classification system. Age ranges for males and females individuals for each of the 3 stages were determined. Stage 3 was observed in only 1 individual (17 years 6 months of age) before the age of 18 years in males, while in the rest of the sample stage 3 was observed in individuals older than 18 years. This suggests a potential value in using stage 3 in determining the age of 18 years in male individuals. The next step was to combine T1W and PDFS MRIs results. This was possible because the analysed T1W MRIs and PDFS MRIs were from the same individuals. Using machine learning algorithms the results obtained using the 2 different MRI settings were combined to verify if the age prediction would have improved and to verify if the final mean absolute error (MAE) would be lower than when using only T1W MRIs. In the female individuals, when adding PDFS MRIs to the T1W MRIs when estimating age, the mean absolute error (MAE) of the age prediction would decrease of 1-3 months, depending on the algorithm used and areas selected for predicting the age. Using only T1W MRIs, the lowest MAE obtained was 12.22 months, while when using a combination of T1W and PDFS, the lowest MAE was 10 months. For the male individuals there was no improvement in the age prediction if PDFS and T1W results were combined, and the lowest MAE for age prediction was 9.33 months in both cases, when only T1W MRIs was used, and T1W and PDFS MRIs were used together. This is probably because female individuals reach maturation earlier than males, and by adding and additional MRIs setting this can assists in providing a better age prediction. The results suggests that for female individuals it would be ideal to use more than one MRIs setting when predicting their age based on maturation of the distal end of the femur.
In conclusion, in the current research, an 8 stages classification system for describing maturation of the distal end of the femur using T1W MRIs was created and tested. Maturation of the growth plate was then described and related to age. Age ranges were provided separately for male and female individuals for each of the classification stages developed. Stages that can be used to determine minimum age of an individual and assist in the age assessment process were also identified. The results obtained were then used to develop automatic age prediction models based on machine learning that can be used to predict the age of those individuals who need their age to be assessed. A classification system based on the analysis of PDFS MRIS was also developed, age ranges for male and female individuals for each stage were provided. The results from the analysis of the T1W and PDFS MRIs were combined, and it was observed that the age prediction improves for the female individuals when both MRI settings are used. It was also observed that stage 3 using the PDFS MRIs could be potentially useful for assessing the minimum age of 18 years in male individuals.
The results of the research will be disseminated via peer reviewed journal articles, reports, and conference presentations.REC name
North of Scotland Research Ethics Committee 1
REC reference
17/NS/0023
Date of REC Opinion
10 Mar 2017
REC opinion
Favourable Opinion