Ref-MIL: Plug-and-Play Posterior Refinement for Multiple Instance Learning in QSM-Based Brain Disease Classification
Authors: Ang Gaoa, Rui Jina, Yin Liu, Pengfei Rong, Feng Liu, G. Bruce Pike, Hongfu Sun, Yang Gao
Organizations: School of Computer Science and Engineering, Central South University, Changsha, China · Department of Radiology, The Third Xiangya Hospital, Central South University, Changsha, China · School of Electrical Engineering and Computer Science, University of Queensland, Brisbane, Australia · Departments of Radiology and Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Calgary, Canada · School of Engineering, University of Newcastle, Newcastle, Australia
Quantitative susceptibility mapping (QSM) is a novel quantitative MRI contrast for studying tissue susceptibility changes in brain disorders, including Parkinson's disease (PD) and multiple sclerosis (MS). However, small cohorts and the lack of local annotations make developing deep learning models for QSM-based diagnosis challenging. In this work, we propose PoRef-MIL, a weakly supervised multiple instance learning (MIL) framework combining iterative slice relabeling and network retraining with a plug-and-play posterior refinement module (PRM). During training, a 2D dual-branch diagnosis network and a subject-level classifier were alternately trained for five rounds. The classifier was used to update slice pseudo-labels for retraining the diagnosis network. The PRM was applied during inference to refine slice probabilities through bidirectional posterior inference and calculate entropy-derived confidence from the refined probabilities, which was then used in sliding-window scoring to obtain subject-level predictions. PoRef-MIL was evaluated on an in-house PD cohort of 287 participants and a public MS cohort of 150 participants. Compared with eight classification methods, PoRef-MIL achieved the highest mean areas under the receiver operating characteristic curve of 0.942 and 0.872, and accuracies of 0.926 and 0.833, for PD and MS classification, respectively. Ablation experiments were conducted to validate the effectiveness of iterative relabeling and PRM, and the plug-and-play feature of the proposed PRM was also evaluated using four state-of-the-art MIL models. Overall, PoRef-MIL offers a promising approach to QSM-based disease classification in small cohorts without manual slice annotations.
Figures & tables
Figure 1: Overview of the 2D dual-branch diagnosis network and its iterative training in PoRef-MIL. (a) In the 2D diagnosis network, each 2D QSM slice is processed by four feature extraction blocks (FEBs), followed by max pooling and slice-level classification. (b) Each FEB block fuses the features from the transformer-based and convolutional branches through cross-attention. (c) Latent features extracted from all 2D slices within a 3D volume are processed by self-attention along the slice direction, and slice scores are then aggregated using attention weights to compute a bag-level focal loss under subject-level supervision. Pseudo-labels derived from these scores are used to retrain the diagnosis network and update the FEB for the next relabeling round. This cycle is repeated for K rounds.
Figure 2: Posterior Refinement Module (PRM) in the proposed PoRef-MIL inference framework. The 3D QSM volumes are treated as a consecutive series of 2D slices, which are fed into the pretrained 2D dual-branch diagnosis network to obtain initial slice probabilities. The slice probabilities and states are then refined using a bidirectional posterior refinement module. After the refinement, entropy-derived confidence is calculated from these probabilities and used in sliding-window scoring to determine the final subject-level predictions.
Characteristics
PD dataset
MS dataset
HC
PD
HC
MS
Participant Counts:
131
156
50
100
Age (years) 1
59.4±8.34
60.7±8.08
57.1±16.7
47.6±11.4
Gender (M/F) 2
55 / 76
81 / 75
19 / 31
42 / 58
Table 1: Demographic characteristics of the PD and MS cohorts
Method
AUC
ACC
SEN
SPE
F1 Score
PD dataset
AG-SE-ResNeXt50 ( Wang et al., 2023 )
0.893±0.002∗∗
0.846±0.015∗∗
0.858±0.018∗∗
0.831±0.021∗∗
0.858±0.014∗∗
MMRN ( Han et al., 2023 )
0.899±0.002∗∗
0.860±0.012∗∗
0.832±0.014∗∗
0.892±0.032∗
0.866±0.010∗∗
CAAE + CRAT ( Cobbinah et al., 2022 )
0.903±0.003∗∗
0.870±0.020∗∗
0.858±0.018∗∗
0.885±0.027∗
0.878±0.019∗∗
DenseFormer-MoE ( Ding et al., 2025 )
0.906±0.001∗∗
0.856±0.015∗∗
0.813±0.014∗∗
0.908±0.021
0.860±0.014∗∗
PMIL ( Yan et al., 2024 )
0.890±0.013∗∗
0.853±0.010∗∗
0.865±0.014∗∗
0.838±0.017∗∗
0.865±0.009∗∗
Table 2: Comparison of the Proposed PoRef-MIL with Eight State-of-the-Art Methods on PD and MS Datasets
Model Configurations
PD dataset
MS dataset ( Fiscone et al., 2024 )
AUC
ACC
F1 Score
AUC
ACC
F1 Score
AttriMIL w/o PRM
0.912±0.002
0.874±0.008
0.884±0.007
0.820±0.010
0.780±0.018
0.840±0.015
AttriMIL + PRM
0.920±0.003
0.888±0.010
0.898±0.008
0.844±0.012
0.807±0.028
0.859±0.021
DA-MIDL w/o PRM
0.914±0.002
0.867±0.010
0.876±0.009
0.802±0.033
0.773±0.043
0.823±0.035
DA-MIDL + PRM
0.924±0.003
0.891±0.015
0.899±0.014
0.834±0.015
0.807±0.028
0.850±0.022
PMIL w/o PRM
0.890±0.013
0.853±0.010
0.865±0.009
0.790±0.021
0.740±0.028
0.810±0.021
Table 3: Plug-and-play Effectiveness of the Proposed Posterior Refinement Module (PRM) Across the Five MIL Models
Figure 3: Comparison of receiver operating characteristic (ROC) curves of the proposed PoRef-MIL with the other methods on PD (a) and MS (b) test sets, with corresponding confusion matrices of PoRef-MIL results (mean ± standard deviation across the five trained models) shown on the right.
Figure 4: Grad-CAM maps and slice-level predictions for representative PD (a, b) and MS (c, d) subjects. (a) and (c) show Grad-CAM maps overlaid on the original QSM slices, with warmer colors indicating stronger contributions. (b) and (d) show the initial slice-level abnormality probability distributions, with red and blue dots indicating slices predicted as abnormal and healthy, respectively. The red and blue dashed step lines represent the corresponding slice states after PRM, with shaded intervals marking slices 45 to 77 in the PD subject and slices 63 to 76 in the MS subject. Purple dashed curves indicate entropy-derived confidence calculated from the refined slice-level probabilities. (e) shows additional coronal and sagittal Grad-CAM heatmaps for a PD subject and an MS subject, respectively, generated using separately trained models.
K
AUC
ACC
F1 Score
PD dataset
– a
0.912±0.002
0.895±0.012
0.903±0.012
0
0.667±0.027
0.639±0.016
0.673±0.012
1
0.913±0.006
0.877±0.018
0.888±0.015
2
0.924±0.005
0.898±0.015
0.906±0.014
3
0.932±0.005
0.909±0.015
0.916±0.013
Table 4: Classification Performance with Conventional MIL and Different Iterative Slice Relabeling Rounds
Inputs
AUC
ACC
F1 Score
PD dataset
T1w
0.908±0.012
0.898±0.015
0.906±0.014
QSM
0.942±0.002
0.926±0.008
0.932±0.008
T1w+QSM
0.964±0.004
0.944±0.015
0.949±0.013
MS dataset
T1w
0.844±0.030
0.800±0.053
0.848±0.040
Table 5: Single-modality and multimodal classification with PoRef-MIL
Figure 5: Comparison of the numbers of model-predicted abnormal slices between healthy controls (HCs) and patients in the (a) PD and (b) MS test sets. Percentages indicate the mean proportion of predicted abnormal slices among all analyzed slices per subject (128 slices for PD and 96 slices for MS).
Age Input
AUC
ACC
F1 Score
PD dataset
w/ FiLM
0.931±0.006
0.916±0.008
0.921±0.008
w/o Age
0.942±0.002
0.926±0.008
0.932±0.008
MS dataset
w/ FiLM
0.860±0.032
0.813±0.061
0.854±0.049
w/o Age
0.872±0.034
0.833±0.053
0.871±0.042
Table 6: Evaluation of FiLM-based Age Conditioning in PoRef-MIL
Paramagnetic rim lesions (Rim+) identified on susceptibility-sensitive MRI have recently emerged as a specific biomarker of chronic active inflammation in Multiple Sclerosis (MS) and are associated with long-term disability progression. However, susceptibility imaging and expert interpretation remain limited to specialized centers, visual assessment is time-consuming and variable, and the low prevalence of Rim+ lesions poses severe class imbalance challenges for automated analysis. We propose a 3D multimodal deep learning framework for lesion-level Rim+/Rim− classification from Quantitative Susceptibility Mapping (QSM) and FLAIR MRI. The architecture explicitly models modality asymmetry by treating QSM as the primary susceptibility-driven signal and conditioning it with FLAIR-derived structural context. To improve robustness under limited data, we employ self-supervised multimodal pretraining followed by supervised fine-tuning with contrastive regularization. The method was evaluated on a clinically acquired cohort of 88 people with MS with expert lesion annotations as reference standard. Results highlight improved performance compared to prior architectures, supporting the effectiveness of asymmetric multimodal modeling for automated chronic active lesion identification.
Introduction: Objective neuroimaging biomarkers may improve Parkinson's disease motor assessment by capturing brain variation not directly observable from clinical examination. We used interpretable machine learning to predict current motor severity, measured by MDS-UPDRS Part III, from QSM and multiband multi-echo resting-state fMRI-derived ReHo features. Methods: Regional QSM and ReHo features were extracted from 28 participants, including 24 individuals with Parkinson's disease and 4 controls. Thirteen feature-set experiments evaluated imaging-only, clinical-only, imaging-plus-clinical, full, reduced, and multimodal inputs. Support vector regression, Elastic Net, Random Forest, and XGBoost models were trained using nested cross-validation. Performance was assessed using pooled held-out R^2, RMSE, MAE, Pearson correlation, permutation testing, and the proportion of participants predicted within +/-5 MDS-UPDRS Part III points. Results: Imaging-only models carried meaningful predictive signal, whereas the clinical-only model performed weakly. Full fMRI, full QSM, and clinical variables provided the strongest global fit, explaining 45.4% of variance in motor severity. Selected QSM plus clinical variables produced the most clinically close predictions, with 75.0% of participants predicted within +/-5 points and the lowest MAE among top-performing models. SHAP highlighted cerebellar, thalamic, striatal, insular, and motor cortical features. Conclusion: QSM and multiband multi-echo fMRI-derived ReHo capture distinct, interpretable dimensions of Parkinson's disease motor severity. These findings show that structural and functional imaging contribute differently depending on the clinical prediction goal.
Aixa X. Andrade
Lyda Hill Department of Bioinformatics · Department of Biomedical Engineering · University of Texas Southwestern Medical Center, Dallas, Texas, USA
Small lesions in brain MRI are hard to segment because they occupy a tiny fraction of the volume and are dominated by background and larger lesions during voxel-wise optimization, so a model can reach a high Dice similarity coefficient (DSC) while missing many of them. We propose CATMIL, a training objective that adds two auxiliary terms to the standard nnU-Net Dice and cross-entropy loss without changing the architecture. The Component-Adaptive Tversky (CAT) term weights lesion voxels by the inverse size of their connected component, so each lesion contributes nearly equally regardless of volume. The lesion-level Multiple Instance Learning (MIL) term treats each lesion as a bag of voxels and penalizes lesions with no detected voxel. For multiple sclerosis lesion segmentation on MSLesSeg, CATMIL achieves the highest small-lesion recall (0.873 vs. 0.796 for Dice+CE; 95% CI of the difference +0.030 to +0.157, higher in all six test patients) and about 48% fewer missed lesions, with comparable DSC and HD95. The gain holds for lesions of at least 3 mm in diameter, the clinical reading size (recall 0.944 vs. 0.870). Standard losses produce no probability response to most small lesions they miss, so no threshold can recover them. The cost is more small false-positive components; a simple component-size filter removes most of them while keeping the sensitivity gain, and at matched lesion-wise precision CATMIL detects more small lesions with higher lesion-wise F1. An ablation attributes the detection gain to the MIL term. On a second dataset, 3D-MR-MS, CATMIL with the same loss weights again improves small-lesion recall, at a larger false-positive cost and slightly lower DSC. Code: https://github.com/luumsk/SmallLesionMRI
Minh Sao Khue Luu, Evgeniy N. Pavlovskiy, Bair N. Tuchinov
The Artificial Intelligence Research Center of Novosibirsk State University, Novosibirsk 630090, Russia