Aim: The aging process differs among the population. To assess an individual’s actual aging degree, researchers have been searching for reliable aging biomarkers. Since brain structure is closely related to aging, it has received much attention. By building brain age estimators using machine learning method, researchers can accurately calculate one’s age, and the predicted age has been found to be able to reflect one’s actual aging degree. However, the past research fails to evaluate models comprehensively. Besides, researchers haven’t selected out a most workable machine learning algorithm for building brain age estimator. Finally, whether should we take age correction remains a question. This study will discuss these two questions, which is the best algorithm and whether should we correct the age bias respectively, by comprehensively comparing models from three aspects of prediction performance, reliability, and validity.
Methods: The three sub-studies of this study discuss these questions respectively from prediction performance, reliability, and validity. Specifically, study 1 builds and trains four models based on four different machine learning algorithms, and then compare their prediction performance using four metrics both before and after age correction. Study 2 calculate the test-retest reliability of four models before and after age correction on datasets containing both pretest and posttest image. Study 3 compares model’s validity by calculating their ability to distinguish patient group from control group and to classify individuals both before and after age correction. Results: Study 1 shows CNN model have the best grade on all four metrics. And age correction can improve the prediction performance of every model. Study 2 shows every model has high reliability and age correction has little impact on it. Study 3 shows CNN model can make best distinguishment and most accurate classification. And age correction can further improve its grade.
Conclusion: Take all three aspects into consideration, CNN is the most competent algorithm for building brain age estimator. Besides, age correction can effectively improve model’s performance. This study can be informative for further research and application of brain age estimator.
修改评论