IC&ML(감진규 교수 연구실), 의료 영상 분야 국제 권위 학회 ‘ISBI 2026’서 논문 3편 발표

• 김에밀 연구원, 최우수 논문에 부여되는 구두 발표 진행

• 학부생 윤혜진 연구원(22학번), 세계적 학술대회서 성과 입증

이미지컴퓨팅및머신러닝연구실(IC&ML lab, 지도교수:감진규) 소속 연구진들이 의료 영상 분석 분야의 세계적 권위 학회인 IEEE ISBI 2026(International Symposium on Biomedical Imaging)에서 총 3편의 연구 논문을 발표했다.

영국 런던에서 개최된 이번 ISBI 2026은 전 세계 의료 AI 전문가들이 모여 최신 기술을 공유하는 자리로, IC&ML 연구실에서는 김에밀, 임영훈, 윤혜진 연구원이 각각 제1저자로 참여한 우수 연구 성과를 선보였다.

특히 김에밀 연구원은 제출 논문 중 상위 소수의 우수 논문에만 부여되는 ‘구두 발표(Oral Presentation)’ 대상자로 선정되어 독보적인 기술력을 입증했다. 또한 윤혜진 연구원(22학번)은 학부생 신분임에도 불구하고 당당히 제1저자로 메이저 국제 학회에서 논문을 발표하며 뛰어난 연구 역량을 증명했다.

각 논문의 상세한 연구 성과는 다음과 같다.

제목: CGM-EEG: Cross-Gated Mamba for Spatio-Temporal EEG Representation Learning

저자: 석박사연계과정 김에밀(제1저자), 감진규(교신저자)

논문 요약: Accurate decoding of electroencephalography (EEG) signals is essential for advancing clinical diagnostics and brain-computer interface (BCI) applications. Although sequence-based models have shown promise in capturing long-range dependencies, they often struggle to represent the intricate spatio-temporal structure of EEG data. To address this limitation, we propose CM-EEG (Cross-Gated Mamba for EEG Decoding), a fully Mamba-based dual-branch architecture that processes spatial and temporal features in parallel. To enhance feature interaction, our framework introduces a Cross-Gate Module (CGM) that enables each branch to integrate information from the other through a lightweight cross-gated mechanism. Extensive experiments on three public clinical EEG benchmarks demonstrate that CGM-EEG achieves up to a 6.6% higher balanced accuracy and a 7.1% lower inference latency than recent transformer-based models. These results highlight the potential of our approach as an efficient and scalable foundation for real-time EEG decoding in biomedical and neuroimaging applications.

제목: CoSMa: Contrastive Learning with Surface Mamba for Infant Brain Age Prediction

저자: 학석사연계과정 윤혜진(제1저자), 감진규(교신저자)

논문 요약: Neonatal cortical analysis demands high precision, yet existing Surface Vision Transformers are constrained by quadratic scaling. We propose CoSMa, a contrastively pre-trained bi-directional surface-based Mamba model for infant brain age prediction. To ensure topological consistency, we project surface features onto an Ico-2 grid and utilize a shared bi-directional Mamba backbone to capture long-range dependencies efficiently. We evaluated CoSMa on a combined dataset (N=884) using 5-fold cross-validation. CoSMa achieves an R-squared of 0.90, significantly outperforming transformer-based and Mamba-based baselines. This demonstrates that contrastive pre-training effectively captures robust global structural representations for accurate infant brain age prediction.

제목: Enhancing Deep Gray Matter Contrast In Atypical Parkinsonism Via SWI-Aware Fusion

저자: 학석사연계과정 임영훈(제1저자), 감진규(교신저자)

논문 요약: Accurate differentiation between Atypical Parkinsonian Syndromes (APS) and Parkinson's Disease (PD) is clinically critical. While Hybrid Contrast (HC) techniques combining T1-weighted (T1w) and Susceptibility-Weighted Imaging (SWI) reflect iron deposition, they often suffer from information loss. We propose an SWI-aware deep learning fusion framework utilizing a Swin Transformer backbone, introducing a novel objective function that integrates an Iron-Preserving Intensity Loss to capture pathological hypointensities and an Artifact-Suppressing Gradient Loss to eliminate boundary artifacts. Validated across three automated segmentation tools (FreeSurfer, FSL, SynthSeg), our framework improved Deep Gray Matter (DGM) visibility and segmentation stability by reducing T1w-driven oversegmentation. Furthermore, downstream analysis demonstrated that our method yields superior AUC and balanced accuracy in APS classification tasks compared to conventional T1-only and HC methods.