Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Jan;3(1):24-33.
doi: 10.47936/encephalitis.2022.00108. Epub 2023 Jan 6.

Improving performance robustness of subject-based brain segmentation software

Jong-Hyeok Park  1 , Kyung-Il Park  2   3 , Dongmin Kim  1 , Myungjae Lee  1 , Shinuk Kang  1 , Seung Joo Kang  4 , Dae Hyun Yoon  5 ; Alzheimer’s Disease Neuroimaging Initiative
Affiliations

Improving performance robustness of subject-based brain segmentation software

Jong-Hyeok Park et al. Encephalitis. 2023 Jan.

Abstract

Purpose: Artificial intelligence (AI)-based image analysis tools to quantify the brain have become commercialized. However, insufficient data for learning and scanner specificity is a limitation for achieving high quality. In the present study, the performance of personalized brain segmentation software when applied to multicenter data using an AI model trained on data from a single institution was improved.

Methods: Preindicators of brain white matter (WM) information from the training dataset were utilized for preprocessing. During learning, data of cognitively normal (CN) individuals from a single center were utilized, and data of CN individuals and Alzheimer disease (AD) patients enrolled in multiple centers were considered the test set.

Results: The preprocessing based on the preindicator (dice similarity coefficient [DSC], 0.8567) resulted in a better performance than without (DSC, 0.7921). The standard deviation (SD) of the WM region intensity (DSC, 0.8303) had a more substantial influence on the performance than the average intensity (DSC, 0.6591). When the SD of the test data WM intensity was smaller than the learning data, the performance improved (0.03 increase in lower SD, 0.05 decrease in higher SD). Furthermore, preindicator-based pretreatment increased the correlation of mean cortical thickness of the entire gray matter between Atroscan and FreeSurfer, and data augmentation without preprocessing did not.Both preindicator processing and data augmentation improved the correlation coefficient from 0.7584 to 0.8165.

Conclusion: Data augmentation and preindicator-based preprocessing of training data can improve the performance of AI-based brain segmentation software, both increasing the generalizability and stability of brain segmentation software.

Keywords: Alzheimer disease; Artificial intelligence; Data augmentation; Segmentation.

PubMed Disclaimer

Conflict of interest statement

Conflicts of Interest Park JH, Kim D, Lee MJ, and Kang S are employees of the JLK. All were blinded to group allocation. Kang SJ, Yoon DH, Lee SK, and Park KI have nothing to disclose. Additional information for correspondence and requests for data should be addressed to Park KI.

Figures

Figure 1.
Figure 1.. Representative images depicting changes due to preprocessing based on signal intensity of white matter
In cases with a small standard deviation (SD) of signal intensity, (A) white matter SD increased and (B) decreased in cases with large SD. (C) Representative figure of cortical thickness measurement. Images have been color-coded for visualization purposes only in this figure. DSC, dice similarity coefficient.
Figure 2.
Figure 2.. The difference in DSC based on white matter preprocessing>
(A) Without augmentation and (B) with augmentation (red line denotes the average standard deviation [SD] of the training set, green circles indicate DSC lower than 0, indicating worse performance by preprocessing). The population with higher SD than the training set showed decreased DSC after preprocessing; augmentation complemented this effect. DSC, dice similarity coefficient.
Figure 3.
Figure 3.. Cortical thickness comparison between AD and CN groups based on index methods
Dice similarity coefficient (A) and mean cortical thickness (B) based on different procedures. Aug, augmentation; Pre, preprocessing; CN, cognitively normal; AD, Alzheimer disease.
Figure 4.
Figure 4.. Correlation of cortical thickness obtained from the developed model with FreeSurfer results
Cortical thickness was compared between Alzheimer disease and cognitively normal groups based on index methods. (A) Original image and no pre-processing [A(−) P(−)]. (B) Original image and pre-processing [A(−) P(+)]. (C) Augmentation and no pre-processing [A(+) P(−)]. (D) Augmentation and pre-processing of white matter [A(+) P(+)]. Freesurfer: version 6.0.0, available at http://surfer.nmr.mgh.harvard.edu.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Alosco ML, Gunstad J, Xu X, et al. The impact of hypertension on cerebral perfusion and cortical thickness in older adults. J Am Soc Hypertens. 2014;8:561–570. - PMC - PubMed
    1. Brundel M, van den Heuvel M, de Bresser J, Kappelle LJ, Biessels GJ, Utrecht Diabetic Encephalopathy Study Group Cerebral cortical thickness in patients with type 2 diabetes. J Neurol Sci. 2010;299:126–130. - PubMed
    1. Ronan L, Alexander-Bloch AF, Wagstyl K, et al. Obesity associated with increased brain age from midlife. Neurobiol Aging. 2016;47:63–70. - PMC - PubMed
    1. Fleischman DA, Arfanakis K, Kelly JF, et al. Regional brain cortical thinning and systemic inflammation in older persons without dementia. J Am Geriatr Soc. 2010;58:1823–1825. - PMC - PubMed
    1. Khedher L, Ramírez J, Górriz JM, Brahim A, Segovia F, Alzheimer’s Disease Neuroimaging Initiative Early diagnosis of Alzheimer’s disease based on partial least squares, principal component analysis and support vector machine using segmented MRI images. Neurocomputing. 2015;151(Part 1):139–150.

LinkOut - more resources