Introduction

 During the Summer 2025 term, I completed a four-month co-op at the University of Guelph Artificial Intelligence (AI) Lab. My work focused on exploring machine learning approaches for medical imaging and fracture risk prediction. At the start of the term, the project centered on identifying datasets suitable for imminent fracture risk prediction in older adults. As the term progressed, we pivoted to the VerSe 2019 dataset, which contains spinal CT scans annotated with vertebrae labels, in order to train models to detect fractures in the thoracic (T) and lumbar (L) regions of the spine. By the end of the work term, I had developed practical experience in medical dataset preparation, machine learning experimentation, and research in a health-focused AI setting.

UofG AI Lab

 The University of Guelph AI Lab is a research group within the School of Computer Science at the University of Guelph. The lab conducts applied and theoretical research in artificial intelligence, with ongoing projects spanning machine learning, natural language processing, computer vision, and health informatics. The lab emphasizes interdisciplinary collaboration, often partnering with healthcare researchers and clinicians to apply AI to real-world challenges.

 Located in Guelph, Ontario, the lab is part of a growing hub of AI research in Canada. Graduate students, faculty, and undergraduate researchers collaborate closely, creating an environment where co-op students are able to contribute meaningfully to ongoing projects. My project, focusing on fracture risk prediction and vertebrae fracture detection, represents the lab’s interest in applying computer vision and machine learning techniques to medical imaging — an area of computing science that has the potential for significant impact on public health.

Goals

Original Goals and Learning Outcomes

 At the beginning of my work term, I set the following goals:

1. Develop proficiency in medical data processing and feature engineering.
2. Build and evaluate predictive models for fracture risk.
3. Explore ground truth strategies for health outcome prediction.
4. Interpret and communicate model insights effectively.
5. Contribute to reproducible and ethical AI research.

 I aimed to strengthen my machine learning research skills while gaining exposure to health-related AI applications. I also hoped to become comfortable working with real-world datasets and frameworks like PyTorch and scikit-learn, as well as tools for reproducibility such as Git and Jupyter notebooks.

Reflection on Goals

• I completed goals relating to dataset processing, model training, and reproducibility. I successfully prepared data pipelines for the VerSe 2019 dataset and trained baseline models to classify fractures in vertebrae.
• The goal of establishing a robust ground truth for imminent fracture risk was not fully achieved, as the datasets available lacked long-term fracture outcomes. This limitation prompted the project pivot.
• I gained experience with PyTorch, medical image preprocessing, and explainable AI techniques, which will directly benefit me in future research or industry co-op roles involving deep learning and healthcare applications.

Job Description

 My primary role was to contribute to research on fracture risk prediction using AI methods. The term began with a focus on surveying and preprocessing datasets that could be used to model imminent fracture risk. This involved data cleaning, exploring variable definitions, and assessing the feasibility of different population health datasets.

 Midway through the term, my work shifted toward computer vision, specifically using the VerSe 2019 dataset of spinal CT scans. My responsibilities included:

• Preprocessing JSON-based vertebrae annotations to extract thoracic and lumbar vertebrae.
• Training classification models to detect fractures in labeled vertebrae.
• Implementing baseline architectures and evaluating them with metrics such as accuracy, sensitivity, and AUC.
• Documenting results and presenting findings to my supervisor.

 This work required skills in Python, PyTorch, and medical image analysis. While I had prior experience with machine learning from coursework, much of the medical imaging knowledge and deep learning experimentation was learned on the job.

Conclusion

 This co-op term provided me with hands-on experience in AI research applied to healthcare. Although the project evolved from its original direction, I gained valuable skills in dataset processing, model training, and reproducibility. More importantly, I learned how to adapt when research goals shift due to data limitations, a common occurrence in applied AI.

 If someone were to summarize my work term, I would want them to say that I contributed meaningfully to an ongoing research project in the University of Guelph AI Lab, gained practical skills in medical AI, and grew as a researcher capable of working at the intersection of computing science and health.