
Contact the VUNO team anytime.
AAIC 2025
Introduction
Cerebral small vessel disease (cSVD) is a progressive neurological condition that impacts brain health through damage to small cerebral blood vessels, predominantly affecting white matter regions. This condition disproportionately affects the elderly population, leading to cognitive decline and elevated stroke risk. The accurate identification of lacunar infarcts has become increasingly crucial for both diagnosing cSVD and determining eligibility for emerging anti-amyloid therapies, as recent studies suggest that the presence of lacunar infarcts may influence treatment response and safety profiles in patients receiving these novel treatments. While brain MRI remains the primary diagnostic tool,
current practice relies heavily on time-intensive radiological interpretation, which is susceptible to inter-reader variability. Recent deep learning approaches for automation face two critical challenges: differentiating lacunes from lacune-mimicking structures (particularly perivascular spaces and vessels) and handling severe data imbalance where normal cases significantly outnumber pathological ones4. To address these challenges, we propose a supervised contrastive learning approach designed to enhance feature discrimination capabilities within imbalanced datasets. This study aimed to develop an automated method that can effectively differentiate lacunes from similar-appearing structures in imbalanced
datasets.
Methods and Materials
Results
We compared three Attention U-Net variants with different encoder initializations: random
initialization, BraTS11,12,13 pre-trained ResNet-34, and our proposed Supervised Contrastive Learning (SupCon) pre-trained ResNet-34 (Table 1). The random initialized model achieved an AUC of 0.732 and average precision of 0.241, while the BraTS11,12,13 pre-trained encoder improved performance with an AUC of 0.747 and average precision of 0.376. The SupCon encoder showed the best performance across all metrics (AUC: 0.804, average precision: 0.504). As shown in Figure 4, GradCAM visualization revealed that SupCon pre-training enables more precise localization of lacunes compared to other methods, demonstrating the importance of encoder pre-training for detecting small, imbalanced lesions in cSVD.
Discussion
In detecting small lesions with severe data imbalance, such as in cSVD, the encoder's discriminative capability becomes crucial. Our performance improvements demonstrate that supervised contrastive learning can effectively enhance the encoder's ability to distinguish subtle features, particularly in separating lacunes from similar-appearing structures. However, important technical limitations remain. The thick-slice nature of 2D FLAIR images (4mm) impedes detection of subtle lacunes, particularly those partially captured between slices, affecting recall performance. Additionally, the severe class imbalance leads to a precision-recall trade-off favoring false positive reduction over complete detection.
Conclusion
This study presents an effective strategy for training encoders to detect small lesions with similar features in cSVD, through supervised contrastive learning. Future work will focus on expanding the current approach to include lacune localization within specific brain regions and external validation across multiple centers.