Endocrine Abstracts

Background: Osteoporosis is a systemic skeletal disease characterized by low bone mass, microarchitectural deterioration of bone tissue, and the consequence of an increased fracture risk. The term “artificial intelligence (AI)” denotes a field in computer science which enables computers to simulate different aspects of human intelligence, such as natural language understanding, pattern recognition or data driven learning. Machine Learning is a subset of AI, and within this field, Convolutional Neural Networks (CNN) play a key role particularly in relation to medical imaging.

Methods: Various AI applications as currently available for the management of osteoporosis are presented, including their strengths and pitfalls.

Results: One of the mainstays of AI supported applications in osteoporosis is imaging based detection of fractures. Almost any medical imaging technique including – but not limited to - plain radiography, computed tomography (CT) and MRT, is represented in an increasing number of studies. There is evidence that AI based technical support can improve fracture detection rate. Furthermore, AI supported algorithms are used not only to assess quantitative aspects of the bone, such as bone mineral density (BMD) at the lumbar spine, the hip and the total body, but also to assess qualitative properties, such as microarchitecture and even fracture load. However, it is of note that many of the clinical studies involving AI in the field of osteoporosis are short of scientific diligence. For example, algorithms behind specific diagnostic approaches are not always published in detail. Also, the logic behind a chosen approach is not always comprehensible. In general, there is a clear lack in standardized procedures.

Conclusion: There are aspects in support of integrating AI based tools into osteoporosis management workflows in daily clinical practice. However, there is also a clear need for high quality clinical research in this field. In this regard, implementation of, e.g., internationally consented quality standards could help to improve the relevance of study outcomes and also their credibility.