# Robust Hierarchical Patterns for identifying MDD patients: A Multisite Study

arXiv: https://arxiv.org/abs/2202.11144 · HTML: https://mlsanity.com/paper/2202.11144/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 92 references that are in the map.
- First posted on arXiv 22 Feb 2022.

## Builds on (most cited first, 12 of 12)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [On the Convergence of Adam and Beyond](https://mlsanity.com/paper/1904.09237/) (2019)
- [A Survey on Multi-Task Learning](https://mlsanity.com/paper/1707.08114/) (2017)
- [Multi-Task Learning as Multi-Objective Optimization](https://mlsanity.com/paper/1810.04650/) (2018)
- [An Overview of Multi-Task Learning in Deep Neural Networks](https://mlsanity.com/paper/1706.05098/) (2017)
- [Pareto Multi-Task Learning](https://mlsanity.com/paper/1912.12854/) (2019)
- [Deriving reproducible biomarkers from multi-site resting-state data: An Autism-based example](https://mlsanity.com/paper/1611.06066/) (2016)
- [A Simple General Approach to Balance Task Difficulty in Multi-Task Learning](https://mlsanity.com/paper/2002.04792/) (2020)
- [Multi-scale detection of hierarchical community architecture in structural and functional brain networks](https://mlsanity.com/paper/1704.05826/) (2017)
- [A Joint Network Optimization Framework to Predict Clinical Severity from Resting State Functional MRI Data](https://mlsanity.com/paper/2009.03238/) (2020)
- [Extraction of Hierarchical Functional Connectivity Components in human brain using Adversarial Learning](https://mlsanity.com/paper/2104.10255/) (2021)
- [Learning Robust Hierarchical Patterns of Human Brain across Many fMRI Studies](https://mlsanity.com/paper/2105.06535/) (2021)

## Abstract (arXiv)

Many supervised machine learning frameworks have been proposed for disease classification using functional magnetic resonance imaging (fMRI) data, producing important biomarkers. More recently, data pooling has flourished, making the result generalizable across a large population. But, this success depends on the population diversity and variability introduced due to the pooling of the data that is not a primary research interest. Here, we look at hierarchical Sparse Connectivity Patterns (hSCPs) as biomarkers for major depressive disorder (MDD). We propose a novel model based on hSCPs to predict MDD patients from functional connectivity matrices extracted from resting-state fMRI data. Our model consists of three coupled terms. The first term decomposes connectivity matrices into hierarchical low-rank sparse components corresponding to synchronous patterns across the human brain. These components are then combined via patient-specific weights capturing heterogeneity in the data. The second term is a classification loss that uses the patient-specific weights to classify MDD patients from healthy ones. Both of these terms are combined with the third term, a robustness loss function to improve the reproducibility of hSCPs. This reduces the variability introduced due to site and population diversity (age and sex) on the predictive accuracy and pattern stability in a large dataset pooled from five different sites. Our results show the impact of diversity on prediction performance. Our model can reduce diversity and improve the predictive and generalizing capability of the components. Finally, our results show that our proposed model can robustly identify clinically relevant patterns characteristic of MDD with high reproducibility.