AI Researcher, Medical Imaging
Samuel Adeniji
Building AI systems for medical image analysis, across 3D CT and MRI as well as vision-language models, with a focus on low-resource clinical settings.
MS Engineering Artificial Intelligence, Carnegie Mellon University
About
Teaching machines to analyze medical images
I work on AI for medical image analysis across 3D CT and MRI. My research covers token-efficient vision-language architectures that let language models reason over CT volumes, along with multi-scale models for tumor classification. I build research systems end to end, from data pipeline through model architecture to evaluation.
As an MS student in Engineering Artificial Intelligence at Carnegie Mellon University, I care most about making diagnostic imaging AI work in low-resource and African clinical settings, which I pursue through the SPARK and CAMERA MRI Africa initiative. My first degree, in Mechatronics Engineering at the Federal University of Technology Minna, still shapes how I approach systems.
Vision-Language Models
Medical Image Analysis
PyTorch
3D CT & MRI
Research
Research Interests
My research asks how AI systems generalize across domains, especially in real-world noisy environments.
Medical Vision-Language Models
Architectures that let language models reason over volumetric medical images, with a focus on token efficiency and how visual representations should be conditioned on the clinical question being asked.
3D Medical Image Analysis
Multi-scale and state-space architectures for CT and MRI, particularly for tumor classification under small, imbalanced datasets where minority-class recall matters clinically.
Efficient & Trustworthy Clinical AI
Model compression and efficiency techniques have real costs in clinical accuracy, calibration, and robustness, and those costs may not fall evenly across patient groups.
AI for Low-Resource Healthcare
Diagnostic imaging AI has to work within the constraints of African and low-resource health systems, including affordable acquisition hardware and limited annotated data.
Projects
Selected Work
ICTC: Instruction-Conditioned Token Compression for 3D CT Vision–Language Models
A dual-stage, instruction-conditioned token compressor for 3D chest CT. A Perceiver stage distils each slice and a FiLM-conditioned stage aggregates slices into a fixed budget of visual tokens for a LoRA-adapted language model that writes the radiology report. On CT-RATE validation (3,039 scans): ROUGE-L 0.273, BLEU-4 0.128, RaTEScore 0.801.
FPN-Mamba: Linear-Time Bidirectional State Space Feature Pyramids for Small-Lesion Brain MRI Classification
Feature pyramid plus bidirectional Mamba for small brain-metastasis screening on UCSF-BMSR. It outperforms a self-attention baseline (InceptentionNet) on AUC (0.982 vs 0.897) and smallest-lesion recall (0.870 vs 0.788) across 5 seeds × 3 folds, and is 3.7× faster per step than a matched attention mixer at 512 px.
Investigating White Blood Cells as a Source of False-Positive Malaria Parasite Detection in African Blood-Smear Images
A controlled study of two YOLOv12s detectors on 8,000 thick-smear images from Uganda and Ghana, using seven spatial and statistical analyses. It finds that white blood cells are not a systematic source of false positives, which redirects mitigation toward staining variability.
Condition-Aware Intensive Care Unit Early-Warning System
Diagnosis-conditioned early-warning models on MIMIC-IV, benchmarking six sequence architectures at 1–3 h horizons. The best model (BiLSTM + attention, ROC-AUC 0.76) is served as a Dockerized FastAPI + WebSocket streaming service, with limitations documented up front.
AI-Powered Reader for Visually Impaired Users
A Raspberry Pi assistive device pairing an HD camera with optical character recognition and speech synthesis, reaching 98% reading accuracy with 40% lower latency. Awarded Best Assistive Project by the Faculty of Engineering, Federal University of Technology, Minna.
Publications
Publications
- 1.
S. A. Adeniji, G. C. Obasi, C.-V. Ntwali, A. M. Iorumbur, C. Raymond, L. Uwimana, A. T. Issah · MICCAI 2026 Workshop: MIRASOL (oral presentation) · 2026
- 2.
S. D. Adetoro, S. A. Adeniji, M. Busogi · NeurIPS 2026 Workshop: Africa in AI · 2026
- 3.
Convergence of Blockchain, Smart Microgrid and Energy Market
K. E. Jack, S. A. Adeniji · Electrica, 24(2), 367–374 · 2024
- 4.
K. E. Jack, S. A. Adeniji · Nigerian Journal of Engineering and Applied Sciences · 2024
- 5.
Design and Implementation of a Secure IoT-Based Remote Monitoring System for an Autonomous Nanny Robot
S. A. Adeniji, K. E. Jack, E. Ani · Submitted, under review
Teaching
Teaching
AI System Design (04-652), Carnegie Mellon University Africa
Fall 2026Teaching Assistant
Supporting a course on edge-to-cloud AI systems, from embedded machine learning on Raspberry Pi and Arduino to cloud-based model analysis.
Cloud Computing (15-319 / 15-619), Carnegie Mellon University
Spring 2026Teaching Assistant
Supported more than 200 undergraduate, master's and doctoral students through office hours, course-overview sessions and grading.
Introduction to Machine Learning, Carnegie Mellon University
Spring 2026Peer Instructor (Volunteer)
Led small-group tutoring on regression, classification, model evaluation and the underlying linear algebra and probability.
Experience
Background
Language and Spatial Laboratories, Carnegie Mellon University
June 2026 – PresentResearch Intern
Research on vision–language models for 3D medical imaging; lead developer of ICTC (see Research Projects). Built the full research pipeline, from data ingestion and the compression projector to multi-task training and evaluation across six visual token budgets.
IEEE SkillUp Hub (Region 8) / Centre for Future Technologies, University of Chichester
Mar 2026 – PresentResearch Assistant
Research on AI-based diagnostic solutions for health systems in low- and middle-income countries.
Artificial Intelligence for Clean Energy Limited, Minna, Nigeria
Oct 2023 – Jan 2025Research Assistant (Intern)
Built a Flask and Random Forest energy-monitoring platform on Google Cloud (99% uptime), improving monitoring accuracy by 30%. Piloted drones for field data collection, cutting collection time by 30% and improving data quality by 12%.
Contact
Let's Connect
Open to research collaborations and roles in AI engineering, particularly in medical imaging or vision-language models. Feel free to reach out.