Codesota · Benchmark · ABIDE IHome/Leaderboards/ABIDE I
Unknown

ABIDE I.

1,112 resting-state fMRI datasets from 539 individuals with autism spectrum disorder (ASD) and 573 typically developing controls across 17 international sites. Multi-site neuroimaging data for autism classification and biomarker discovery.

Paper ↗Leaderboard ↓
§ 01 · SOTA history

Year over year.

§ 02 · Leaderboard

Results by metric.

Found a wrong score or missing run?
Use row edits to send a sourced correction into moderation.
Add / edit result ↗Report issue ↗

accuracy

Accuracy is the reported evaluation metric for ABIDE I. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for accuracyverifiedpapervendorcommunityunverified

Muted rows were not state of the art when published — an earlier or same-year result already scored better.

RankModelTrustScoreYearLinksFix
01SSAE + Softmax (Explainable ASD)
DPARSF pipeline, 5-fold CV, 884 subjects with FD < 0.2mm head movement filtering. F1=0.97, Sensitivity=0.98, Specificity=0.98, Precision=0.98. State-of-the-art on ABIDE I.
verified98.22025Paper ↗Source ↗Looks wrong?
02Plymouth DL Model
98% on 884 participant subset. Highlights visual cortex regions.
paper982025Source ↗Looks wrong?
03plymouth-dl-model
98% on 884 participant subset. Highlights visual cortex regions.
paper982025Source ↗Looks wrong?
04MCBERT
Multi-modal CNN-BERT with leave-one-site-out cross-validation. Combines brain MRI and meta-features.
paper93.42025Source ↗Looks wrong?
05ae-fcn
Autoencoder + FCN combining fMRI and sMRI data (Rakic et al., 2020).
paper852025Source ↗Looks wrong?
06Multi-Atlas DNN
Multi-atlas deep neural network with hinge loss. 79.13% on augmented data.
paper78.072025Source ↗Looks wrong?
07multi-atlas-dnn
Multi-atlas deep neural network with hinge loss. 79.13% on augmented data.
paper78.072025Source ↗Looks wrong?
08asd-swnet
Shared-weight feature extraction network. Precision: 76.15%, Recall: 80.65%.
paper76.522025Source ↗Looks wrong?
09MSalNET
Multi-site adversarial learning with node information assembly. Site-level adversarial learning to mitigate confounding site effects. Also evaluated on ADHD-200.
verified75.562025Paper ↗Looks wrong?
10MADE-for-ASD
Full ABIDE I dataset (1,035 subjects). Sensitivity: 82.90%, Specificity: 69.70%. 96.40% on NYU subset. Improves 4.4pp over prior works.
verified75.22024Paper ↗Looks wrong?
11maacnn
Multi-attention CNN. AUC: 0.79 on ABIDE-I.
paper75.122025Source ↗Looks wrong?
12ChebGAT-GCN
Chebyshev Spectral GCN + GAT multi-branch architecture for multimodal neuroimaging. Outperforms conventional GCNs, autoencoders, and multimodal CNNs on full ABIDE I.
verified74.822025Paper ↗Looks wrong?
13al-negat
Adversarial learning-based node-edge graph attention network. 1,007 subjects across 17 sites.
paper74.72025Source ↗Looks wrong?
14ASDFormer
Transformer with Mixture of Pooling-Classifier Experts (MoE). Sensitivity: 82.55% ±10.19, Specificity: 66.09% ±4.74. AUC: 81.17%.
verified74.62025Paper ↗Looks wrong?
15RGTNet
AAL atlas, 5-fold cross-validation. Residual Graph Transformer with Graph Sparse Fitting. Outperforms competitors at 70.9%.
verified73.42024Source ↗Looks wrong?
16BrainGNN
ROI-aware graph convolutional network. Interpretable biomarker discovery. 1,035 subjects.
paper73.32025Source ↗Looks wrong?
17gcn
Graph Convolutional Network combining fMRI and sMRI with max voting.
paper72.22025Source ↗Looks wrong?
18Multi-Task Transformer
Multi-task transformer neural network on UM dataset. Attention mechanism for feature extraction.
paper722025Source ↗Looks wrong?
19multi-task-transformer
Multi-task transformer neural network on UM dataset. Attention mechanism for feature extraction.
paper722025Source ↗Looks wrong?
20Causal fMRI Model
Causality-inspired deep learning on fMRI time-series. Identifies left/right precuneus and cerebellum as top causal ROIs. Interpretable ASD classification.
verified71.92025Paper ↗Looks wrong?
21phgcl-ddgformer
Graph contrastive learning with graph transformer. 74.8% sensitivity.
paper70.92025Source ↗Looks wrong?
22SVM with Connectivity Features
Support Vector Machine with functional connectivity features. Classic baseline comparison.
paper70.12025Source ↗Looks wrong?
23svm-connectivity
Support Vector Machine with functional connectivity features. Classic baseline comparison.
paper70.12025Source ↗Looks wrong?
24BrainTWT
Temporal random walk + transformer dynamic network embedding. Sensitivity: 65.04%, Specificity: 74.35%. 2.38% relative improvement over GSA-LSTM.
verified70.032025Paper ↗Looks wrong?
25deep-learning-heinsfeld
Deep learning approach by Heinsfeld et al. (2017). Anterior-posterior brain connectivity disruption.
paper702025Source ↗Looks wrong?
26Deep Learning (Heinsfeld)
Deep learning approach by Heinsfeld et al. (2017). Anterior-posterior brain connectivity disruption.
paper702025Source ↗Looks wrong?
27MVS-GCN
Multi-view Site Graph Convolutional Network handling multi-site variability.
paper69.382025Source ↗Looks wrong?
28abraham-connectomes
Participant-specific functional connectivity matrices (Abraham et al., 2017).
paper672025Source ↗Looks wrong?
29Abraham Connectomes
Participant-specific functional connectivity matrices (Abraham et al., 2017).
paper672025Source ↗Looks wrong?
30random-forest
Random Forest baseline. Sensitivity: 69%, Specificity: 58%.
paper632025Source ↗Looks wrong?
31Random Forest
Random Forest baseline. Sensitivity: 69%, Specificity: 58%.
paper632025Source ↗Looks wrong?

auc

Auc is the reported evaluation metric for ABIDE I. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for aucverifiedpapervendorcommunityunverified

Muted rows were not state of the art when published — an earlier or same-year result already scored better.

RankModelTrustScoreYearLinksFix
01ChebGAT-GCN
AUC of 0.82 on full ABIDE I dataset with multimodal Chebyshev+GAT approach.
verified822025Paper ↗Looks wrong?
02ASDFormer
AUC-ROC. Outperforms Com-BrainTF (78.77%) and BrainNetTF (77.58%).
verified81.172025Paper ↗Looks wrong?
03ASD-SWNet
AUC-ROC score for autism vs typical control classification.
paper812025Source ↗Looks wrong?
04braingт
Graph Transformer for brain disorder diagnosis. Significantly outperforms BrainNetTF (73.2%).
paper78.72025Source ↗Looks wrong?
05BrainGT
Graph Transformer for brain disorder diagnosis. Significantly outperforms BrainNetTF (73.2%).
paper78.72025Source ↗Looks wrong?
06gcn
Best performing model in comprehensive comparison study (2024).
paper782025Source ↗Looks wrong?
07SVM with Connectivity Features
Traditional ML baseline with functional connectivity matrices.
paper772025Source ↗Looks wrong?
08svm-connectivity
Traditional ML baseline with functional connectivity matrices.
paper772025Source ↗Looks wrong?
09Causal fMRI Model
AUC-ROC. Causal modeling with inter-ROI causality analysis.
verified75.82025Paper ↗Looks wrong?
10BrainTWT
AUC-ROC. 6.77% relative improvement over GSA-LSTM in AUC.
verified75.272025Paper ↗Looks wrong?
11MVS-GCN
Multi-view approach addressing site heterogeneity.
paper69.012025Source ↗Looks wrong?
§ 04 · Submit a result

Add to the leaderboard.

← Back to Leaderboards