Research internship experience in Medical AI and Computer Vision at GIST
I participated in an undergraduate research internship at GIST from 2025 to 2026. During this period, I conducted research in Medical AI and Computer Vision across the AI-MED (AI on Medical Applications) Lab and the DMCB (Data Mining and Computational Biology) Lab.
My primary research focus was on predicting spatial transcriptomics from pathology images. In simple terms, this task involves estimating the spatial distribution of gene expression levels of cells directly from histopathology images.
Through this experience, I gained hands-on expertise in deep learning, data analysis, and interdisciplinary research. In particular, I studied a wide range of deep learning paradigms, including Convolutional and Transformer-based vision encoders, genetic language models, foundation models, self-supervised learning, multimodal representation learning, and contrastive learning. By actively participating in weekly lab meetings and iteratively developing research ideas, I strengthened my ability to think critically, tackle complex problems with persistence, and effectively communicate technical concepts with others.
Overall, this period marked a significant phase of both academic and personal growth.
At the GIST AI-MED Lab (Sep 2025 – Present), I have been conducting research in Medical AI with a focus on computational pathology and multimodal representation learning. I developed RaPaCL-ST (RadiomicsFeature-Pathomics Contrastive Learning), a multimodal framework designed for spatial transcriptomics prediction. This work aimed to bridge handcrafted radiomics features and deep pathomics features extracted from histopathology images. By leveraging contrastive learning, I aligned interpretable radiomics signals—such as texture and heterogeneity—with high-dimensional image embeddings in a shared latent space, improving both the biological relevance and interpretability of learned representations for downstream tasks like spatial gene expression prediction. Through this work, I gained hands-on experience in model design, multimodal learning, and research-driven problem solving.
Previously, at the GIST DMCB Lab (Mar 2025 – Aug 2025), I participated in the G-SURF (GIST Summer Undergraduate Research Fellowship) program, where I conducted research on extending foundation models for single-cell epigenomics to neurodegenerative diseases. I proposed a method to adapt EpiAgent, an in-silico perturbation framework originally developed for cancer, to Alzheimer’s disease (AD). To address brain cell heterogeneity, I designed a framework incorporating cell-type specific LoRA (Low-Rank Adapters), enabling more precise modeling of diverse cell populations such as neurons, astrocytes, microglia, and oligodendrocytes. Applying this approach to single-cell ATAC-seq datasets from AD and control brain tissues, I improved the model’s ability to simulate regulatory responses across different cell types. This experience strengthened my skills in applying foundation models, handling high-dimensional biological data, and designing domain-specific model extensions.
Here are some more articles you might like to read next: