Toufiq Musah

MSE Data Science Student

University of Pennsylvania

News

Presenting at the COMPASS Webinar on Automated 3D Medical Image Segmentation, Sept 2025.

Won the Brain Tumor Segmentation Challenge at MICCAI in Daejeon, Sep 2025.
Presenting at the MIRASOL and Deep-Breath Workshop at MICCAI in Daejeon, Sep 2025.
Attending the African Computer Vision Summer School in Kigali, Jul 2025.
Hosting a tutorial on Optimisation Techniques at Ghana Data Science Summit , Jun 2025.

About Me

I am an MSE Data Science Student at the University of Pennsylvania, with interests in Artificial Intelligence Applications in Health and Medicine. My research spans Computer Vision, Medical Image Analysis, Multi-Omic Data Integration, and Trustworthy AI for Healthcare Delivery, Developing Reliable and Interpretable Clinical Decision Support Tools.

Research Interests

Computer Vision
Medical Image Analysis
Artificial Intelligence in Medicine

Education

BSc Biomedical Engineering

Jan 2021 - Nov 2024

Kwame Nkrumah University of Science and Technology (KNUST)

MSE Data Science

Aug 2025 - Present

University of Pennsylvania (UPenn)

Experience

Research Engineer

Kumasi Centre for Collaborative Research in Tropical Medicine • Oct 2024 - Oct 2025

Research Assistant

Responsible Artificial Intelligence Lab (RAIL) • Oct 2023 - Feb 2024

Selected Publications

Towards Trustworthy Breast Tumor Segmentation in Ultrasound using Monte Carlo Dropout and Deep Ensembles for Epistemic Uncertainty Estimation

Toufiq Musah, Chinasa Kalaiwo, Maimoona Akram, Ubaida Napari Abdulai, Maruf Adewole, Farouk Dako, Adaobi Chiazor Emegoakor, Udunna C Anazodo, Prince Ebenezer Adjei, Confidence Raymond, Medical Image Computing in Resource Constrained Settings Workshop. 28th International MICCAI Conference, (2025), arXiv:2508.17768.

[arXiv]

How We Won BraTS-SSA 2025: Brain Tumor Segmentation in the Sub-Saharan African Population Using Segmentation-Aware Data Augmentation and Model Ensembling

Claudia Takyi Ankomah, Livingstone Eli Ayivor, Ireneaus Nyame, Leslie Wambo, Patrick Yeboah Bonsu, Aondona Moses Iorumbur, Raymond Confidence, Toufiq Musah, Brain Tumor Segmentation, and Cross-Modality Domain Adaptation for Medical Image Segmentation. 28th International MICCAI Conference, (2025), arXiv:2510.03568.

[arXiv]

Automated Segmentation of Ischemic Stroke Lesions in Non-Contrast Computed Tomography Images for Enhanced Treatment and Prognosis

Toufiq Musah, Prince Ebenezer Adjei, Kojo Obed Otoo (2024). Medical Image Computing and Computer Assisted Interventions 2240, 9.

[Springer] [arXiv]

Large Kernel MedNeXt for Breast Tumor Segmentation and Self-Normalizing Network for pCR Classification in Magnetic Resonance Images

Toufiq Musah. Artificial Intelligence and Imaging for Diagnostic and Treatment Challenges in Breast Care: Second Deep Breast Workshop, Deep-Breath (2025)

[Springer] [arXiv]

Brain Tumor Segmentation in Sub-Sahara Africa with Advanced Transformer and ConvNet Methods: Fine-Tuning, Data Mixing and Ensembling

Toufiq Musah, C Amoako-Atta, JA Otu, LE Ismaila, SA Suraka, O Williams, ... Medical Image Computing in Resource Constrained Settings Workshop. 28th International MICCAI Conference, (2025), arXiv:2508.10905.

[arXiv]

Projects

MedSegMNIST

A Python library for easy access to 2D and 3D Medical Image Segmentation datasets provided in varying resolutions and sizes, designed for benchmarking and learning purposes.

Multi-Omic Data Integration for Head and Neck Tumor Prognostication

Multimodal learning framework to predict 5-year survivability and recurrence rates. Integrated whole-slide image features from CLAM with structured clinical data, laboratory analytes, surgical notes, and patient history. Used ClinicalBERT for text embeddings and low-rank bilinear fusion, achieving an AUC of 0.75 across tasks.

Vision-Language Surgical Scene Understanding & Panoptic Segmentation

End-to-end system integrating surgical video data and panoptic segmentations for automated operating room scene understanding. Trained S2-Scaled SwinUNETR for panoptic segmentation across 21 classes. Fine-tuned MedGemma using QLoRA for scene graph generation and event significance detection.

Collaborative AI Resources for Expectant Mothers (C.A.R.E)

Accessible AI assistant built on Large Language Models with Retrieval-Augmented Generation to help expectant mothers with gestational diabetes. Deployed through WhatsApp for inclusivity and ease of access, providing personalized health guidance in low-resource settings.

Generative AI for Medical Imaging Enhancement and Data Augmentation

Developed generative models to improve medical imaging quality and data efficiency. Super-resolution and denoising framework for high-resolution head CT scans from low-resolution variants, reducing radiation exposure. Implemented Deep Convolutional GANs for synthetic medical image generation for self-supervised pre-training and data-efficient fine-tuning.