Interpretable Deep Learning for Neuroimaging-Based Diagnostic Classification

被引:2
作者
Deshpande, Gopikrishna [1 ,2 ,3 ,4 ,5 ,6 ]
Masood, Janzaib [1 ]
Huynh, Nguyen [1 ]
Denney Jr, Thomas S. [1 ,2 ,3 ,4 ]
Dretsch, Michael N. [7 ]
机构
[1] Auburn Univ, Neuroimaging Ctr, Dept Elect & Comp Engn, Auburn, AL 36849 USA
[2] Auburn Univ, Dept Psychol Sci, Auburn, AL 36849 USA
[3] Alabama Adv Imaging Consortium, Birmingham, AL 35294 USA
[4] Auburn Univ, Ctr Neurosci, Auburn, AL 36849 USA
[5] Natl Inst Mental Hlth & Neurosci, Dept Psychiat, Bengaluru 560029, India
[6] Indian Inst Technol Hyderabad, Dept Heritage Sci & Technol, Hyderabad 502285, India
[7] Joint Base Lewis McChord, Walter Reed Army Inst Res West, West, WA 98433 USA
关键词
Resting-state functional magnetic resonance; resting-state functional connectivity; interpretable deep learning; POSTTRAUMATIC-STRESS-DISORDER; ANTERIOR CINGULATE CORTEX; RESTING-STATE FMRI; FUNCTIONAL CONNECTIVITY; NETWORKS; ABUSE; MEMORIES; VETERANS; DISEASE; SERVICE;
D O I
10.1109/ACCESS.2024.3388911
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Deep neural networks (DNN) are increasingly being used in neuroimaging research for the diagnosis of brain disorders and understanding of human brain. Despite their impressive performance, their usage in medical applications will be limited unless there is more transparency on how these algorithms arrive at their decisions. We address this issue in the current report. A DNN classifier was trained to discriminate between healthy subjects and those with posttraumatic stress disorder (PTSD) using brain connectivity obtained from functional magnetic resonance imaging data. The classifier provided 90% accuracy. Brain connectivity features important for classification were generated for a pool of test subjects and permutation testing was used to identify significantly discriminative connections. Such heatmaps of significant paths were generated from 10 different interpretability algorithms based on variants of layer-wise relevance and gradient attribution methods. Since different interpretability algorithms make different assumptions about the data and model, their explanations had both commonalities and differences. Therefore, we developed a consensus across interpretability methods, which aligned well with the existing knowledge about brain alterations underlying PTSD. The confident identification of more than 20 regions, acknowledged for their relevance to PTSD in prior studies, was achieved with a voting score exceeding 8 and a family-wise correction threshold below 0.05. Our work illustrates how robustness and physiological plausibility of explanations can be achieved in interpreting classifications obtained from DNNs in diagnostic neuroimaging applications by evaluating convergence across methods. This will be crucial for trust in AI-based medical diagnostics in the future.
引用
收藏
页码:55474 / 55490
页数:17
相关论文
共 50 条
[41]   Development of a Bispectral index score prediction model based on an interpretable deep learning algorithm [J].
Hwang, Eugene ;
Park, Hee -Sun ;
Kim, Hyun-Seok ;
Kim, Jin-Young ;
Jeong, Hanseok ;
Kim, Junetae ;
Kim, Sung-Hoon .
ARTIFICIAL INTELLIGENCE IN MEDICINE, 2023, 143
[42]   Multiphysical Interpretable Deep Learning Network for Oil Spill Identification Based on SAR Images [J].
Fan, Jianchao ;
Sui, Zitai ;
Wang, Xinzhe .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2024, 62 :1-15
[43]   Interpretable deep learning: interpretation, interpretability, trustworthiness, and beyond [J].
Xuhong Li ;
Haoyi Xiong ;
Xingjian Li ;
Xuanyu Wu ;
Xiao Zhang ;
Ji Liu ;
Jiang Bian ;
Dejing Dou .
Knowledge and Information Systems, 2022, 64 :3197-3234
[44]   A model-based deep learning approach to interpretable impact force localization and reconstruction [J].
Zhou, Rui ;
Qiao, Baijie ;
Jiang, Liangliang ;
Cheng, Wei ;
Yang, Xiuyue ;
Wang, Yanan ;
Chen, Xuefeng .
MECHANICAL SYSTEMS AND SIGNAL PROCESSING, 2025, 224
[45]   Interpretable deep learning: interpretation, interpretability, trustworthiness, and beyond [J].
Li, Xuhong ;
Xiong, Haoyi ;
Li, Xingjian ;
Wu, Xuanyu ;
Zhang, Xiao ;
Liu, Ji ;
Bian, Jiang ;
Dou, Dejing .
KNOWLEDGE AND INFORMATION SYSTEMS, 2022, 64 (12) :3197-3234
[46]   Fully interpretable deep learning model of transcriptional control [J].
Liu, Yi ;
Barr, Kenneth ;
Reinitz, John .
BIOINFORMATICS, 2020, 36 :499-507
[47]   Industry return prediction via interpretable deep learning [J].
Zografopoulos, Lazaros ;
Iannino, Maria Chiara ;
Psaradellis, Ioannis ;
Sermpinis, Georgios .
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH, 2025, 321 (01) :257-268
[48]   Radioport: a radiomics-reporting network for interpretable deep learning in BI-RADS classification of mammographic calcification [J].
Pang, Ting ;
Wong, Jeannie Hsiu Ding ;
Ng, Wei Lin ;
Chan, Chee Seng ;
Wang, Chang ;
Zhou, Xuezhi ;
Yu, Yi .
PHYSICS IN MEDICINE AND BIOLOGY, 2024, 69 (06)
[49]   From estimating activation locality to predicting disorder: A review of pattern recognition for neuroimaging-based psychiatric diagnostics [J].
Wolfers, Thomas ;
Buitelaar, Jan K. ;
Beckmann, Christian F. ;
Franke, Barbara ;
Marquan, Andre F. .
NEUROSCIENCE AND BIOBEHAVIORAL REVIEWS, 2015, 57 :328-349
[50]   Classification of Neurodegenerative Diseases Based on VGG 19 Deep Transfer Learning Architecture: A Deep Learning Approach [J].
Bhatele, Kirti Raj ;
Bhadauria, Sarita Singh .
BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS, 2020, 13 (04) :1972-1980