Biology subjects

Zemlyanker, D.

Publications and source records attributed to Zemlyanker, D..

3 recordsLinked to original sources

bioRxiv2026

Accessing Mesoscale Functional Organization from Standard Functional MRI Leveraging AI-Based Super-Resolution

Access to the fine-scale functional organization of the human brain is essential for understanding cognition and neurological disorders yet remains beyond the reach of conventional functional MRI (fMRI) because of its limited spatial resolution and contrast-to-noise ratio. To overcome this limitation, we developed f-LENNS (functional Localization Enhancement via Neural-Net-Sharpening), an artificial intelligence (AI)-based framework that enhances the effective spatial resolution of reconstructed fMRI magnitude images. The model was trained exclusively on high-resolution synthetic MRI data using extensive domain randomization and no adversarial or generative components. This approach enables robust out-of-the-box generalization (without retraining) to experimental datasets obtained with diverse acquisitions, while effectively curbing hallucination. Our experiments show that our publicly available model substantially improves the spatial fidelity of functional activation maps across multiple imaging paradigms and generalizes to a cohort of individuals with atypical brain function caused by amblyopia (lazy eye). F-LENNS consistently outperforms conventional resolution enhancement approaches, including interpolation-based upsampling and sharpening methods. These findings demonstrate that AI-driven super-resolution combined with domain randomization can overcome domain shifts arising from differences in MRI acquisition, enabling scalable, fine-scale functional brain mapping with widely available MRI hardware.

neuroscience↗

bioRxiv2026

Improving racial fairness in brain age models using style-transfer synthesis

Brain age models can systematically mispredict age for specific demographic groups, risking biased estimates of neurological health. We present an approach to improve accuracy and reduce racial disparities using synthetic T1 images generated via style-transfer with SuperSynth, an open-source FreeSurfer tool that produces intensity harmonized isotropic images regardless of the input's contrast or resolution. We refer to the original scans as the real domain and their SuperSynth-derived counterparts as the synthetic domain. Because each synthetic image derives from the participant's own scan, comparisons between domains hold anatomy and demographic composition constant. We audited fairness by training a neural network across thirteen racial compositions on a diverse cohort (683 White, 605 Black, 431 Asian) in both domains. Our findings highlight three key insights. First, the synthetic domain enhanced both fairness and accuracy, and even models trained on a single demographic showed reduced racial disparity purely from switching domain. The same pattern appeared in two independently developed pretrained models. Second, models trained on synthetic data demonstrated superior out-of-distribution robustness in external clinical testing. Third, augmenting imbalanced datasets with synthetic minority images closed 61% of the fairness gap without requiring new data collection, and unlike explicitly supplying race labels, does not require race as a model input at deployment. Representational analyses indicated that appearance standardization changes which features drive age prediction rather than removing demographic information from the images. By utilizing synthetic data, our approach offers a robust pathway to more equitable and generalizable brain age models.

bioengineering↗

bioRxiv2026

On the accuracy of image registration in portable low-field 3D brain MRI

Portable low-field MRI offers an affordable and mobile alternative to conventional high-field scanners, enabling imaging in point-of-care and resource-limited settings. However, its lower signal-to-noise ratio, reduced resolution, and acquisition artifacts raise concerns about the accuracy of standard image registration methods. Reliable registration is critical for a wide range of emerging applications, including frequent brain monitoring, assessment of neurodegenerative disease progression, and evaluation of treatment effects such as those of Alzheimers therapeutics. In this work, we systematically evaluated state-of-the-art registration approaches on simulated low-field scans (obtained by downsampling high-field images) and on real low-field brain MRI data. We compared three representative approaches: classical optimization (NiftyReg), learning-based registration (SynthMorph), and synthesis-based registration (SynthSR+NiftyReg). Using downsampled high-field scans, all methods performed well, achieving high Dice scores and smooth deformation fields, indicating that reduced resolution alone does not hinder registration. In contrast, real low-field data exhibited lower accuracy, primarily due to geometric distortion and other acquisition-specific artifacts. Among the tested approaches, the synthesis-based pipeline achieved the most robust performance across subjects and modalities. Overall, existing algorithms can accommodate resolution limitations, however, future methods could further enhance coregistration by explicitly addressing the distortions present in low-field MRI scans.

neuroscience↗