Research of diffusion tensor imaging autism spectrum disorder classification model based on the visual transformer deep learning algorithm

doi:10.3877/cma.j.issn.1673-5250.2025.01.006

Objective

To explore the clinical screening efficacy of children with autism spectrum disorder（ASD）in different ages by ASD classification model based on the visual transformer（ViT）deep learning algorithm diffusion tensor imaging （DTI）（hereinafter referred to as the"ASD classification model"）.

Methods

A total of 265 ASD children aged 0-12 years who visited the Chinese PLA General Hospital and the Chengdu Women's and Children's Central Hospital from January 2017 to July 2023 were selected into ASD group，And other 158 typical development（TD）children aged 0-12 years old who visited the above two hospitals during the same period were included in the control group.Brain DTIimages of all children were collected，and the DTIimage data of 34 and 64 directions of children from the two hospitals were harmonized，and then whole brain white matter fibers of the children were traced based on two-tensor unscented Kalman filter （t UKF）technology.Based on spectral embedding technology，fiber level TractoEmbedding images of the children were generated.Then augmented TractoEmbedding images were classified using ViT deep learning algorithm，and ASD classification model was established，and the screening performance of the model for ASD with different ages（0-2 years old，≥2-4 years old，≥4-6 years old，≥6-8 years old，≥8-10 years old，≥10-12 years old）was analyzed，including accuracy，specificity，sensitivity，and area under the curve（AUC）.This study was approved by the Medical Ethics Committee of Chinese PLA General Hospital（Approval No.S2022-646-01）.All guardians of the enrolled children signed the informed consent form for clinical study.There was no statistically s ignificant difference in the age of children aged≥2-4 years，≥4-6 years，≥6-8 years，≥8-10 years，and≥10-12 years between ASD group and control group （P＞0.05）.

Results

Based on the ViT deep learning algorithm，this study established five ASD classification models （Models 1-5）by training and validating fiber level DTI TractoEmbedding images of ASD and TD children.Among them，Model 3 had the highest screening accuracy for ASD at 0.791.The results of screening performance in children of different ages by Model 3 showed that the accuracy，specificity，sensitivity，and AUC values of the model for ASD screening in children aged≥4-5 years，≥5-6 years，≥6-7 years，and≥7-8 years were as follows.≥4-5 years old：0.966，0.864，1.000，0.961；≥5-6 years old：0.891，0.842，0.926，0.961；≥6-7 years old：0.958，0.903，1.000，0.992；≥7-8 years old：0.893，0.850，1.000，0.956.

Conclusions

The ASD classification model 3 based on ViT deep learning algorithm DTI established in this study has high sensitivity，specificity，accuracy，and AUC values for ASD screening in children aged≥4-5 years，≥5-6 years，≥6-7 years，and≥7-8 years.

Key words: Autism spectrum disorder, Diffusion tensor imaging, Whole brain white matter fiber, Vision transformer, Neurodevelopmental disorders, Child

图1 基于ViT 深度学习算法DTI的ASD 分类模型建立与验证流程图注：ViT 为视觉转换器，DTI为弥散张量成像，ASD 为孤独症谱系障碍。t UKF为双张量无迹卡尔曼滤波，Deep WMA 为深层白质分析，FA 为各向异性分数。ORG 图谱为O'Donnell研究组纤维聚类白质图谱（O'Donnell Research Group Fiber Clustering White Matter Atlas）。TractoEmbedding图像为本研究基于t UKF技术对受试儿脑白质纤维示踪后，应用谱嵌入技术降维生成的二维图像

表1 ASD 组与对照组不同年龄段受试儿的年龄与性别比较

组别	0～2岁			≥2～4岁			≥4～6岁
组别	例数	年龄（岁， $\overset{ˉ}{x}$ ± s ）	男性［例数（%）］	例数	年龄［岁，M ( Q₁,　 Q₃ )或 $\overset{ˉ}{x}$ ± s ］	男性［例数（%）］	例数	年龄［岁，M ( Q₁,　 Q₃ )］	男性［例数（%）］
ASD组	2	1.7±0.2	1（50.0）	82	3.3（3.0，3.6）	66（80.5）	121	4.8（4.3，5.3）	95（78.5）
对照组	0	-	-	2	3.4±0.2^a	1（50.0）	60	4.8（4.3，5.4）	34（56.7）
统计量		-	-		Z=-0.177	χ ²=1.12		Z=-0.26	χ ²=9.35
P值		-	-		0.860	0.289		0.799	0.002
组别	≥6～8岁			≥8～10岁			≥10～12岁
组别	例数	年龄［岁，M ( Q₁,　 Q₃ )］	男性［例数（%）］	例数	年龄（岁， $\overset{ˉ}{x}$ ± s ）	男性［例数（%）］	例数	年龄［岁，M ( Q₁,　 Q₃ )］	男性［例数（%）］
ASD组	49	6.4（6.1，6.8）	36（73.5）	10	8.8±0.6	9（90.0）	1	10.6	1（100.0）
对照组	51	6.5（6.2，7.2）	36（70.6）	38	8.7±0.5	19（50.0）	7	10.8（10.1，10.9）	4（57.1）
统计量		Z=-1.58	χ ²=0.10		t=0.75	χ ²=5.21		Z=0.44	χ ²=0.69
P值		0.114	0.748		0.458	0.022		0.657	0.408

表1 ASD 组与对照组不同年龄段受试儿的年龄与性别比较

组别	0～2岁			≥2～4岁			≥4～6岁
组别	例数	年龄（岁， $\overset{ˉ}{x}$ ± s ）	男性［例数（%）］	例数	年龄［岁，M ( Q₁,　 Q₃ )或 $\overset{ˉ}{x}$ ± s ］	男性［例数（%）］	例数	年龄［岁，M ( Q₁,　 Q₃ )］	男性［例数（%）］
ASD组	2	1.7±0.2	1（50.0）	82	3.3（3.0，3.6）	66（80.5）	121	4.8（4.3，5.3）	95（78.5）
对照组	0	-	-	2	3.4±0.2^a	1（50.0）	60	4.8（4.3，5.4）	34（56.7）
统计量		-	-		Z=-0.177	χ ²=1.12		Z=-0.26	χ ²=9.35
P值		-	-		0.860	0.289		0.799	0.002
组别	≥6～8岁			≥8～10岁			≥10～12岁
组别	例数	年龄［岁，M ( Q₁,　 Q₃ )］	男性［例数（%）］	例数	年龄（岁， $\overset{ˉ}{x}$ ± s ）	男性［例数（%）］	例数	年龄［岁，M ( Q₁,　 Q₃ )］	男性［例数（%）］
ASD组	49	6.4（6.1，6.8）	36（73.5）	10	8.8±0.6	9（90.0）	1	10.6	1（100.0）
对照组	51	6.5（6.2，7.2）	36（70.6）	38	8.7±0.5	19（50.0）	7	10.8（10.1，10.9）	4（57.1）
统计量		Z=-1.58	χ ²=0.10		t=0.75	χ ²=5.21		Z=0.44	χ ²=0.69
P值		0.114	0.748		0.458	0.022		0.657	0.408

图2 基于ViT 深度学习算法DTI的ASD 分类模型3的TractoEmbedding图像分类注意力图（图2A～2C：TractoEmbedding图像嵌入分辨率分别为320、160、80的左脑TractoEmbedding图像分类注意力图；图2D～2F：TractoEmbedding图像嵌入分辨率分别为320、160、80的右脑TractoEmbedding图像分类注意力图；图2H～2J：TractoEmbedding图像嵌入分辨率分别为320、160、80的左、右脑联合区TractoEmbedding图像分类注意力图）注：橙色所示为ASD分类模型3对受试儿TractoEmbedding图像的关注区域。ViT 为视觉转换器，DTI为弥散张量成像，ASD 为孤独症谱系障碍。TractoEmbedding图像为本研究基于双张量无迹卡尔曼滤波技术对受试儿脑白质纤维示踪后，应用谱嵌入技术降维生成的二维图像

图3 5个基于ViT 深度学习算法DTI 的ASD 分类模型中训练集与验证集受试儿的年龄、性别分布图（图3A：模型1中训练集与验证集受试儿的年龄、性别分布图；图3B：模型2中训练集与验证集受试儿的年龄、性别分布图；图3C：模型3中训练集与验证集受试儿的年龄、性别分布图；图3D：模型4中训练集与验证集受试儿的年龄、性别分布图；图3E：模型5中训练集与验证集受试儿的年龄、性别分布图）注：ViT 为视觉转换器，DTI为弥散张量成像，ASD为孤独症谱系障碍

表2 ASD 分类模型3对不同年龄段受试儿ASD 筛查效能比较

不同年龄段受试儿	ASD患儿例数	TD儿童例数	总例数	准确率	特异度	敏感度	AUC值
0～2岁	2	0	2	-	-	-	-
≥2～4岁	82	2	84	1.000	1.000	1.000	1.000
≥4～6岁	121	60	181	0.920	0.850	0.967	0.968
≥6～8岁	49	51	100	0.940	0.882	1.000	0.981
≥8～10岁	10	38	48	0.896	0.868	1.000	1.000
≥10～12岁	1	7	8	1.000	1.000	1.000	1.000

表2 ASD 分类模型3对不同年龄段受试儿ASD 筛查效能比较

不同年龄段受试儿	ASD患儿例数	TD儿童例数	总例数	准确率	特异度	敏感度	AUC值
0～2岁	2	0	2	-	-	-	-
≥2～4岁	82	2	84	1.000	1.000	1.000	1.000
≥4～6岁	121	60	181	0.920	0.850	0.967	0.968
≥6～8岁	49	51	100	0.940	0.882	1.000	0.981
≥8～10岁	10	38	48	0.896	0.868	1.000	1.000
≥10～12岁	1	7	8	1.000	1.000	1.000	1.000

表3 ASD 分类模型3对进一步细分的不同年龄段（年龄间隔为1岁）受试儿ASD 筛查效能比较

不同年龄段受试儿	ASD患儿例数	TD儿童例数	总例数	准确率	特异度	敏感度	AUC值
≥4～5岁	67	32	89	0.966	0.864	1.000	0.961
≥5～6岁	54	38	92	0.891	0.842	0.926	0.961
≥6～7岁	41	31	72	0.958	0.903	1.000	0.992
≥7～8岁	8	20	28	0.893	0.850	1.000	0.981
≥8～9岁	6	25	31	0.903	0.880	1.000	0.956
≥9～10岁	4	13	17	1.000	1.000	1.000	1.000

表3 ASD 分类模型3对进一步细分的不同年龄段（年龄间隔为1岁）受试儿ASD 筛查效能比较

不同年龄段受试儿	ASD患儿例数	TD儿童例数	总例数	准确率	特异度	敏感度	AUC值
≥4～5岁	67	32	89	0.966	0.864	1.000	0.961
≥5～6岁	54	38	92	0.891	0.842	0.926	0.961
≥6～7岁	41	31	72	0.958	0.903	1.000	0.992
≥7～8岁	8	20	28	0.893	0.850	1.000	0.981
≥8～9岁	6	25	31	0.903	0.880	1.000	0.956
≥9～10岁	4	13	17	1.000	1.000	1.000	1.000

［1］	郝伟，陆林.精神病学［M］.8版.北京：人民卫生出版社，2018：228-231.Hao W，Lu L.Psychiatry［M］.8th.Beijing：People's Medical Publishing House，2018：228-231.
［2］	Baio J，Wiggins L，Christensen DL，et al.Prevalence of autism spectrum disorder among children aged 8 yearsautism and developmental disabilities monitoring network，11 sites，United States，2014［J］.MMWR Surveill Summ，2018，67（6）：1-23.DOI：10.15585/mmwr.ss6706a1.
［3］	Khachadourian V，Mahjani B，Sandin S，et al.Comorbidities in autism spectrum disorder and their etiologies［J］.Transl Psychiatry，2023，13（1）：71.DOI：10.1038/s41398-023-02374-w.
［4］	Li SJ，Wang Y，Qian L，et al.Alterations of white matter connectivity in preschool children with autism spectrum disorder［J］.Radiology，2018，288（1）：209-217.DOI：10.1148/radiol.2018170059.
［5］	Yeh CH，Tseng RY，Ni HC，et al.White matter microstructural and morphometric alterations in autism：implications for intellectual capabilities［J］.Mol Autism，2022，13（1）：21.DOI：10.1186/s13229-022-00499-1.
［6］	Wolff JJ，Gu H，Gerig G，et al.Differences in white matter fiber tract development present from 6 to 24 months in infants with autism［J］.Am J Psychiatry，2012，169（6）：589-600.DOI：10.1176/appi.ajp.2011.11091447.
［7］	Han K，Wang Y，Chen H，et al.A survey on vision transformer［J］.IEEE Trans Pattern Anal Mach Intell，2023，45（1）：87-110.DOI：10.1109/TPAMI.2022.315 2247.
［8］	Thompson A，Shahidiani A，Fritz A，et al.Age-related differences in white matter diffusion measures in autism spectrum condition［J］.Mol Autism，2020，11（1）：36.DOI：10.1186/s13229-020-00325-6.
［9］	López-Vicente M，Lamballais S，Louwen S，et al.White matter microstructure correlates of age，sex，handedness and motor ability in a population-based sample of 3031 school-age children［J］.Neuroimage，2021，227：117643.DOI：10.1016/j.neuroimage.2020.117643.
［10］	毕小彬，范晓壮，米文丽，等.ICD-11和DSM-5中孤独症谱系障碍诊断标准比较［J］.国际精神病学杂志，2021，48（2）：193-196.Bi XB，Fan XZ，Mi WL，et al.A comparative analysis of diagnostic criteria for autism spectrum disorder in ICD-11 and DSM-5［J］.J Int Psychiat，2021，48（2）：193-196.
［11］	Oguz I，Farzinfar M，Matsui J，et al.DTIPrep：quality control of diffusion-weighted images［J］.Front Neuroinform，2014，8：4.DOI：10.3389/fninf.2014.00004.
［12］	Mirzaalian H，Ning L，Savadjiev P，et al.Multi-site harmonization of diffusion MRI data in a registration framework［J］.Brain Imaging Behav，2018，12（1）：284-295.DOI：10.1007/s11682-016-9670-y.
［13］	Malcolm JG，Shenton ME，Rathi Y.Filtered multitensor tractography［J］.IEEE Trans Med Imaging，2010，29（9）：1664-1675.DOI：10.1109/TMI.2010.2048121.
［14］	Norton I，Essayed WI，Zhang F，et al.Slicer DMRI：open source diffusion mri software for brain cancer research［J］.Cancer Res，2017，77（21）：e101-e103.DOI：10.1158/0008-5472.CAN-17-0332.
［15］	Zhang F，Wu Y，Norton I，et al.An anatomically curated fiber clustering white matter atlas for consistent white matter tract parcellation across the lifespan［J］.Neuroimage，2018，179：429-447.DOI：10.1016/j.neuroimage.2018.06.027.
［16］	O'Donnell LJ，Wells WM，Golby AJ，et al.Unbiased groupwise registration of white matter tractography［J］.Med Image Comput Comput Assist Interv，2012，15（Pt 3）：123-130.DOI：10.1007/978-3-642-33454-2_16.
［17］	O'Donnell LJ，Westin CF.Automatic tractography segmentation using a high-dimensional white matter atlas［J］.IEEE Trans Med Imaging，2007，26（11）：1562-1575.DOI：10.1109/TMI.2007.906785.
［18］	Johnston MV，Nishimura A，Harum K，et al.Sculpting the developing brain［J］.Adv Pediatr，2001，48：1-38.
［19］	Johnston MV.Clinical disorders of brain plasticity［J］.Brain Dev，2004，26（2）：73-80.DOI：10.1016/S0387-7604（03）00102-5.
［20］	Di Martino A，O'Connor D，Chen B，et al.Enhancing studies of the connectome in autism using the autism brain imaging data exchangeⅡ［J］.Scientific Data，2017，4：170010.DOI：10.1038/sdata.2017.10.
［21］	Helmy E，Elnakib A，El Nakieb Y，et al.Role of artificial intelligence for autism diagnosis using dti and fmri：a survey［J］.Biomedicines，2023，11（7）：1858.DOI：10.3390/biomedicines11071858.
［22］	Payabvash S，Palacios EM，Owen JP，et al.White matter connectome edge density in children with autism spectrum disorders：potential imaging biomarkers using machinelearning models［J］.Brain Connect，2019，9（2）：209-220.DOI：10.1089/brain.2018.0658.
［23］	Cetin Karayumak S，Bouix S，Ning L，et al.Retrospective harmonization of multi-site diffusion MRI data acquired with different acquisition parameters［J］.Neuroimage，2019，184：1800-200.DOI：10.1016/j.neuroimage.2018.08.073.
［24］	Malcolm JG，Michailovich O，Bouix S，et al.A filtered approach to neural tractography using the Watson directional function［J］.Med Image Anal，2010，14（1）：58-69.DOI：10.1016/j.media.2009.10.003.
［25］	Baumgartner C，Michailovich O，Levitt J，et al.A unified tractography framework for comparing diffusion models on clinical scans［C］//CDMRI-workshop（MICCAI'12），2012.
［26］	He J，Zhang F，Xie G，et al.Comparison of multiple tractography methods for reconstruction of the retinogeniculate visual pathway using diffusion MRI［J］.Hum Brain Mapp，2021，42（12）：3887-3904.DOI：10.1002/hbm.25472.
［27］	Vercruysse D，Christiaens D，Maes F，et al.Fiber bundle segmentation using spectral embedding and supervised learning［C］//O'Donnell L，Nedjati-Gilani G，Rathi Y，et al.Computational diffusion MRI.Mathematics and visualization.Berlin：Springer，2014，39：103-114.DOI：10.1007/978-3-319-11182-7_10.
［28］	Khan A，Rauf Z，Sohail A，et al.A survey of the vision transformers and their CNN-transformer based variants［J］.Artif Intell Rev，2023，56.DOI：10.1007/s10462-023-10595-0.
［29］	Dosovitskiy A，Beyer L，Kolesnikov A，et al.An image is worth 16x16 words：transformers for image recognition at scale ［C ］//International Conference on Learning Representations，2021.
［30］	Calderoni S.Sex/gender differences in children with autism spectrum disorder：A brief overview on epidemiology，symptom profile，and neuroanatomy［J］.J Neurosci Res，2023，101（5）：739-750.DOI：10.1002/jnr.25000.
［31］	Randall M，Egberts KJ，Samtani A，et al.Diagnostic tests for autism spectrum disorder（ASD）in preschool children［J］.Cochrane Database Syst Rev，2018，7（7）：CD009044.DOI：10.1002/14651858.CD009044.pub2.

[1]	Meijuan Gao, Suzhen Dong. Current research status of multimodal imaging in pediatric common abdominal malignant tumors[J]. Chinese Journal of Obstetrics & Gynecology and Pediatrics(Electronic Edition), 2025, 21(01): 15-20.
[2]	Litong Ni, Shijun Li. Study on amygdala volume and amygdala white matter fiber connection in preschool children with autism spectrum disorder[J]. Chinese Journal of Obstetrics & Gynecology and Pediatrics(Electronic Edition), 2025, 21(01): 29-36.
[3]	Yuchun Yan, Xinyu Yuan, Zhaohui Xu, Chunmei Zhu, Yang Yang. Impact of chest CT air bronchogram on decision-making for bronchoalveolar lavage in children with Mycoplasma pneumoniae pneumonia[J]. Chinese Journal of Obstetrics & Gynecology and Pediatrics(Electronic Edition), 2025, 21(01): 54-60.
[4]	Tianli Niu, Chunyan Niu, Pengfei Zhao, Shijun Li. Clinical value of arterial spin labeling in detecting cerebral blood flow in children with pilocytic astrocytoma，hemangioblastoma and medulloblastoma[J]. Chinese Journal of Obstetrics & Gynecology and Pediatrics(Electronic Edition), 2025, 21(01): 61-66.
[5]	Huashan Zhao, Weiguo Rong, Yunpeng Zhai, Rui Guo, Hongxiu Xu, Sai Huang, Gang Shen, Shisong Zhang. Postoperative retained appendicoliths causing umbilical incision infection in children with acute appendicitis：a case report and literature review[J]. Chinese Journal of Obstetrics & Gynecology and Pediatrics(Electronic Edition), 2025, 21(01): 92-97.
[6]	Medical Quality Center Central for Infectious Diseases National. Expert consensus on quality improvement for prevention of mother-to-child transmission of hepatitis B virus[J]. Chinese Journal of Experimental and Clinical Infectious Diseases(Electronic Edition), 2025, 19(01): 23-29.
[7]	Mengjie Jiang, Shujuan Huang, Yuxin Pei, Liping Rong, Yuanyuan Xu, Zhilang Lin, Yuanquan Qiu, Longshan Liu, Xiaoyun Jiang, Lizhi Chen. Clinical features and treatment of recurrent allograft nephropathy in children[J]. Chinese Journal of Transplantation(Electronic Edition), 2025, 19(01): 16-21.
[8]	Xiaoyan Jin, Fang Deng, Shaohan Fang, Yating Ding. Analysis of cardiac structure and function in pediatric kidney transplant recipients[J]. Chinese Journal of Transplantation(Electronic Edition), 2025, 19(01): 26-30.
[9]	Zhixin Liu, Xiaolin Cheng, Wenjun Xiao, Siqi Liu. Application of day surgery mode in laparoscopy for hydrocele in children[J]. Chinese Journal of Endourology(Electronic Edition), 2025, 19(01): 36-40.
[10]	Lijie Fan, Yali Wei. Effect of S(+)-ketamine and sufentanil on intraoperative hemodynamics and postoperative spontaneous recovery in children undergoing laparoscopic high ligation of hernia sac[J]. Chinese Journal of Hernia and Abdominal Wall Surgery(Electronic Edition), 2025, 19(02): 214-218.
[11]	Kuilin Lyu, Hong Chen. Analysis of epidemiological and clinical characteristics of 465 pediatric cases of Mycoplasma pneumoniae pneumonia： a single-center retrospective study[J]. Chinese Journal of Lung Diseases(Electronic Edition), 2025, 18(01): 55-61.
[12]	Wenhai Bao, Qin Ruan, Feifei Li. Application value of Th17/Treg immune imbalance in the diagnosis and disease evaluation of children with abdominal Henoch-Schonlein purpura[J]. Chinese Journal of Digestion and Medical Imageology(Electronic Edition), 2025, 15(02): 162-167.
[13]	Caiyun Han, Jiao Ma, Yan Li. Study on the correlation between serum PAF and sTM levels and rotavirus enteritis and dehydration degree in children[J]. Chinese Journal of Digestion and Medical Imageology(Electronic Edition), 2025, 15(01): 83-88.
[14]	Yun Wang, Lei Liu, Min Xu, De Huai. Clinical efficacy and safety of dupilumab injection combined with Yupingfeng granules in treatment of moderate to severe atopic dermatitis in children[J]. Chinese Journal of Clinicians(Electronic Edition), 2025, 19(01): 27-32.
[15]	Rongchang Wu, Xiaolian Fang, Lei Yang, Chengyue Zhang, Junyang Zhao, Yun Peng, Jia You, Jie Yin. Complications of intra-arterial chemotherapy for retinoblastomas：a single-center study analysis[J]. Chinese Journal of Interventional Radiology(Electronic Edition), 2025, 13(01): 49-53.

Viewed

Full text

Abstract

Please choose a citation manager

Content to export