Bao Ngoc Huynh's doctoral work shows that deep learning can provide automatic diagnostics in both human and veterinary medicine that is both fast and interpretable for clinicians.
The focus of Huynh’s doctoral thesis at the Norwegian University of Life Sciences (NMBU) was the application of deep learning to medical diagnostics. The amount of medical data collected during patient examinations is growing rapidly, and healthcare professionals are struggling to keep up with and process this vast amount of information. Deep learning could assist them by automating tasks like medical image analysis and prediction of patient outcomes, allowing doctors to focus on more complex cases and providing better care for patients.
Huynh's research explored several clinical tasks that can be time-consuming, labor-intensive and potentially affected by slight differences in diagnostics between clinicians. The cases included automatic segmentation of cancer tumors for radiotherapy planning in both humans and canines, prediction of cancer treatment outcomes, and diagnostics of canine elbow dysplasia.
The findings emphasized the potential of deep learning to significantly enhance healthcare by providing faster diagnoses, improving treatment planning, and reducing discrepancies in interpretations among doctors.
Huynh's research, along with other studies, demonstrates that deep learning models can achieve performance levels comparable to human experts, potentially leading to more efficient and consistent diagnoses.
There are, however, challenges in applying deep learning to healthcare, particularly the risk of significant errors and the need to improve model transparency and reliability to build trust among medical professionals. Other challenges, such as small dataset sizes, data variability, and low data quality, were also addressed in Huynh's thesis.
Huynh investigated various strategies to enhance model performance and trustworthiness, paving the way for their use in real-world clinical practice. These strategies included various techniques for data preprocessing and augmentation, transfer learning, explainable artificial intelligence, and quality assurance methods.
Huynh’s research holds promise for advancing academic knowledge and addressing real-world challenges. It contributes to our understanding of how to use deep learning for medical image analysis, showing the strengths and limitations of these models and how to improve them further. In real-world healthcare, these advancements could automate tasks and improve diagnostic accuracy, leading to better and faster treatment for patients.
However, while the evidence supports the potential of deep learning, more research is needed before these models can be widely implemented in hospitals. It is important to remember that deep learning is a tool designed to assist health professionals, not replace them.
Bao Ngoc Huynh defends her PhD thesis "Applications of deep learning for medical data analysis" on Friday 8 November 2024. Read more about the event here.