The doctoral dissertation in the field of Technical Physics will be examined at the Faculty of Science, Forestry and Technology, Kuopio campus and online.
What is the topic of your doctoral research? Why is it important to study the topic?
My doctoral research investigated how computational methods could support the diagnosis of osteoporosis and improve the assessment of fracture risk. As populations age, osteoporosis and the resulting fractures are becoming increasingly common, challenging the healthcare system.
Osteoporosis is diagnosed by measuring bone mineral density (BMD) using dual-energy X-ray absorptiometry (DXA). Although BMD predicts fractures at the population level, its ability to assess an individual's risk is limited. My thesis investigates whether computational methods, including deep learning, can extract additional predictive information from DXA images and other medical imaging data.
What are the key findings or observations of your doctoral research?
The results suggest that convolutional neural networks can detect pathological findings in DXA images but capture only limited additional information for fracture prediction. This supports the current understanding that fracture risk depends on many interacting factors, only some of which are visible in medical images.
Another focus of the thesis was the detection of existing fragility fractures, which serve not only as the outcome of interest in research, but also as one of the strongest predictors of future fracture risk. It was found that fragility fracture data in medical registers can be complemented by algorithmic analysis of radiological examination patterns.
The thesis also developed a deep learning-based image analysis method for detecting vertebral compression fractures. The results indicate potential for opportunistic screening and retrospective identification of vertebral fractures from multiple imaging modalities.
How can the results of your doctoral research be utilised in practice?
The deep learning models developed for detecting clinical findings, such as scoliosis and degenerative changes, could support the interpretation of DXA scans by alerting clinicians to potential sources of inaccuracy. These findings are common and may reduce the accuracy and consistency of bone density measurements.
The fracture detection methods proposed in the dissertation can improve the identification of prior fractures, which are strong predictors of future fracture risk. This could support screening strategies aimed at earlier identification of high-risk patients and enable timely prevention.
The study findings also improve understanding of fracture data recorded in Finnish healthcare systems regarding completeness and accuracy. The developed methods enable efficient collection of fracture data from large healthcare datasets and could therefore support future fracture research.
What are the key research methods and materials used in your doctoral research?
During the doctoral research, a retrospective dataset of nearly 90 000 patients who underwent DXA scans at Finnish hospitals was acquired. In addition to DXA images, the dataset included images from other modalities, as well as national register data on the patients’ healthcare visits and diagnoses. The dataset was used in the thesis and could also support future research on osteoporotic fractures.
Deep learning methods, specifically convolutional neural networks, were used for image analysis tasks in multiple imaging modalities. In supervised learning, the models are trained on input images paired with corresponding output labels, which allows them to learn the feature extraction capabilities needed in the task.
Rule-based algorithms were also developed to identify fracture events. In addition to national registers, the algorithms use information about X-ray examination visits recorded in medical image archives.
The doctoral dissertation of Tomi Nissinen, MSc, entitled Detection and prediction of osteoporotic fractures from medical imaging data will be examined at the Faculty of Science, Forestry and Technology, Kuopio campus. The opponent will be Assistant Professor Danielle Whittier, University of Calgary, Canada, and the custos will be Docent Sami Väänänen, University of Eastern Finland. Language of the public defence is English.
For further information, please contact:
Tomi Nissinen, [email protected]
