‘Deep learning algorithm’ in orthopaedic trauma

‘Artificial intelligence’ and ‘deep learning’ are the way of the future when it comes to medicine.

Already they are showing promising results as valuable diagnostic tools to help clinicians across many specialties, such as survival estimation in patients with bone tumours.

However, when it comes to caring for and treating bone fractures, these tools aren’t used by surgeons in daily practice.

With the help of a Flinders Foundation Health Seed Grant, Professor Gregory Bain, a hand and upper limb surgeon at Flinders Medical Centre and Professor of Upper Limb and Research at Flinders University, will look to develop and test a deep learning algorithm to assist in orthopaedic trauma by detecting, classifying and characterising proximal humerus fractures on CT scans and radiographic imaging views.

Proximal humerus fractures often occur from falling with an outstretched arm.

“A well performing algorithm is quick, can assist in surgical decisions regarding whether to operate or not operate, and can assess images consistently without fatigue or distractions in the context of emergency care trauma management,” Professor Bain explains.

The deep learning algorithm will classify proximal humerus fractures according to the 14 basic fracture types, and works using pattern recognition to analyse pixels which humans are not able to detect. The accuracy of the algorithm will then be compared against the traditional performance of surgeons.

“The algorithm can potentially enhance workflow and provide a clinical tool to improve diagnostic assessment in addition to scans …this means that in the future, surgeons may consult the algorithm if they have doubts about the presence of a fracture or want confirmation,” Professor Bain says.

“It will lead to improved surgical decision making as the algorithm will be trained by highly experienced surgeons and will therefore perform as good as they are, and other young surgeons can use this algorithm to assist in surgical decision making based on fracture characteristics.

“The algorithm is intended to neutralize the influence of surgeons and limit misdiagnosis.”


Research category: Orthopaedic Trauma

Project title: Development of Deep Learning Algorithms for Prediction of Fracture Detection, Classification and Characteristics for Proximal Humerus Fractures: Does the Computer Outperform Surgeons?

Lead researcher: Professor Gregory Bain

Support seed funding research projects at Flinders DONATE NOW