Hey there! 👋 I’m a first, second third year Ph.D. candidate. in Neuroscience at McGill University 🇨🇦, proudly affiliated with the Montreal Neurological Institute. Under the guidance of Dr. Amir Shmuel and Dr. Janine Mendola, my research lives at the exciting crossroads of Artificial Intelligence, Medical Imaging, and Neuroscience.

Research Interests 💻🧠: I’m currently working on applying computer vision to several medical imaging downstream tasks such as registration, segmentation, and modality translation, using state-of-the-art neural networks, multimodal architectures, and vision-language models (VLMs).

I’m open to new ideas and collaboration. Please feel free to contact me at gurucharan dot marthikrishnakumar at mail dot mcgill dot ca.

📧 Email  /  📄 CV  /  💼 LinkedIn  /  💻 GitHub  /  📚 Google Scholar

I’ve explored the applications of AI to medical imaging, working with structural MRI 🧠, diffusion MRI 💧, and tractography 🧬. My past work includes image registration, segmentation, and retrieval, as well as developing multimodal and vision-language models. I’ve published on these topics and continue to work on advancing deep learning methods for medical image analysis. Below are a few of my selected publications:

Gurucharan Marthi Krishna Kumar, Aman Chadha, Janine Mendola, Amir Shmuel

ICCV Workshop on Computer Vision for Automated Medical Diagnosis (CVAMD), Oct 2025

🧠 Medical Image Segmentation | Code available
MedVisionLlama combines vision transformers with pre-trained language model layers to boost medical image segmentation performance, especially when training data is scarce. Using Low-Rank Adaptation (LoRA) for efficient fine-tuning, it performs well on MRI, CT, and other scans. The model shows sharper focus on important image regions and produces more accurate, stable results compared to standard approaches.

Gurucharan Marthi Krishna Kumar, Janine Mendola, Amir Shmuel

IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025, Mar 2025

🧠 Medical Image Registration | Code available
NestedMorph is a novel deep learning method that aligns brain MRI scans by combining fine image details and broader context using nested attention layers. This helps the model achieve highly accurate alignment of complex brain structures. Tested on a large dataset, it showed strong improvements in accuracy and reliability.

Gurucharan Marthi Krishna Kumar, Vijay Jeyakumar, Siddarth S

Book Chapter, Nov 2024 (Undergraduate Work)

🧠 Medical Image Retrieval
This project created a deep learning system that helps find medical images quickly by combining 2D images like X-rays and 3D scans like MRI and CT. It uses several neural networks to learn important features and find similar cases faster ⚡. The best model achieved very high accuracy, supporting quicker and better clinical decisions.