An MRI-Based Interpretable Deep Learning Model for AD Risk Screening and Progression Prediction
Presented During:
Wednesday, June 25, 2025: 5:45 PM - 7:00 PM
Brisbane Convention & Exhibition Centre
Room:
M3 (Mezzanine Level)
Poster No:
130
Submission Type:
Abstract Submission
Authors:
Bin Lu1, Yan-Rong Chen1, Rui-Xian Li2, Ming-Kai Zhang2, Shao-Zhen Yan2, Min Wei2, Francisco Castellanos3, Paul Thompson4, Jie Lu2, Ying Han2, Chao-gan Yan5
Institutions:
1Institute of Psychology, Chinese Academy of Sciences, Beijing, Beijing, 2Xuanwu Hospital, Capital Medical University, Beijing, Beijing, 3New York University School of Medicine, New York, NY, 4University of Southern California, Los Angeles, CA, 5Tsinghua University, Beijing, China
First Author:
Co-Author(s):
Introduction:
Alzheimer's disease (AD) is a progressive neurodegenerative disease that poses a significant challenge to global health, profoundly impacting individuals, families, and healthcare systems. Early detection of AD is crucial, as it allows for timely interventions that could slow disease progression and improve patient outcomes. The advent of recent novel immunotherapies has further heightened the need for cost-effective and time-efficient biomarkers for early diagnosis to enhance treatment effectiveness(Hansson, 2021). Deep learning methods has revolutionized MRI's potential towards earlier and more precise AD screening and facilitating timely medical treatment. In addition, interpretable models could enhance the confidence of physicians and patients in medical imaging models.
Methods:
The current study included 722 participants from a longitudinal dataset (SILCODE, 60.80% female; 66.10 ± 7.70 years of age at baseline), categorized based on clinical assessments into 211 having normal cognition (NC), 235 with subjective cognitive decline (SCD), 193 with Mild cognitive impairment (MCI), and 83 AD participants at baseline, with the longest follow-up period reaching up to 14 years (Li et al., 2019). The participants completed 1,980 visits and provided 1,105 valid T1-weighted MRI scans. A generalizable AD classification model was pretrained based on large and diverse structural MRI datasets, encompassing over 85,000 samples from 217 different sites/scanners in our previous study (Lu et al., 2022). The model achieved an accuracy rate exceeding 90% across four independent AD datasets and was directly applied to the SILCODE cohort without any re-training or fine-tuning. We tested the model's ability in AD classification, progression prediction, and correlating with plasma biomarkers and cognitive measurements. Additionally, we calculated individualized AD brain risk maps and subtyped and predicted the speed of cognitive decline via the interpretable model.
Results:
The classifier, which had been trained on North American samples, demonstrated good generalizability in distinguishing AD from NC within the Chinese sample, achieving an AUC of 91.3%, an accuracy of 88.2%, a sensitivity of 95.2%, and a specificity of 85.6% (Figure 1a). MCI patients defined as high risk by the model were significantly more likely to progress to AD. The median survival time (i.e., time-to-conversion) of the high-risk MCI group (31 months) was significantly shorter than that of the low-risk MCI group (>167 months) (Figure 1c). Importantly, the model accurately identified 86.7% of participants who would progress to AD up to 14 years before progression. The classifier achieved a capture rate about 90% within 3 years, 78% within 3–5 years, and 57% within 5–14 years before the first AD diagnosis among AD converters (Figure 1d). The MRI-based risk scores were significantly correlated with global cognitive measurements and plasma-based biomarkers such as p-tau 217 and NfL (Figure 1e-f). The interpretable model provided personalized brain risk maps for every single patient. MCI patients are divided into three subtypes based on individualized brain risk maps (Figure 2a-d), one of which is prone to rapid cognitive decline (Figure 2e-f).
·Prediction Performance of the AD classfier.
·Subtyping and Predicting Cognitive Decline Speed via an Interpretable Model.
Conclusions:
In summary, our MRI-based deep learning AD classifier generalized to a different population, identifying 87% of participants at baseline who progressed to AD. The interpretable models indicated AD risk brain regions and predicted the MCI patients who experienced the fastest cognitive declines. Implementation of MRI in clinical practice can complement current biomarkers such as CSF, PET, and blood tests, facilitating the selection of patients most likely to benefit from prospective monitoring and early immunotherapy interventions, thereby supporting clinical decision-making and precision medicine.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Keywords:
Other - Alzheimer’s disease; Deep learning; Progression prediction; Interpretable Model, Plasma biomarkers
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Li, X., Wang, X., Su, L., Hu, X., & Han, Y. (2019). Sino Longitudinal Study on Cognitive Decline (SILCODE): Protocol for a Chinese longitudinal observational study to develop risk prediction models of conversion to mild cognitive impairment in individuals with subjective cognitive decline. BMJ Open, 9(7), e028188. https://doi.org/10.1136/bmjopen-2018-028188
Lu, B., Li, H.-X., Chang, Z.-K., Li, L., Chen, N.-X., Zhu, Z.-C., Zhou, H.-X., Li, X.-Y., Wang, Y.-W., Cui, S.-X., Deng, Z.-Y., Fan, Z., Yang, H., Chen, X., Thompson, P. M., Castellanos, F. X., & Yan, C.-G. (2022). A practical Alzheimer’s disease classifier via brain imaging-based deep learning on 85,721 samples. Journal of Big Data, 9(1), 101. https://doi.org/10.1186/s40537-022-00650-y