By: 31 January 2022
How new developments in AI are accurately detecting fractures in X-rays

In a busy emergency room where medical attention must be divided among patients, X-rays are key to determining case severity.

However, depending on the time of day or night a patient presents, the number of specialists available, and factoring elements such as fatigue, number of X-rays analysed that day, and importance of the case, it can be relatively easy to miss a fracture diagnosis on X-ray.

New developments in artificial intelligence (AI) are making it possible to diagnose fractures more accurately than ever.

Gold standard imaging modality for diagnosing a fracture

When patients present for the investigation of a fracture, healthcare professionals first prefer to understand the incident leading to injury.

Fractures, based on age, can occur for a variety of reasons.

Younger individuals may document recent injuries resulting in a fracture, while elderly patients might present at a much later stage following single or multiple fractures while documenting negligible trauma. This is primarily due to the increased risk of falls and a high incidence of osteoporosis within this demographic.

In either case, the gold standard imaging modality to detect and diagnose a fracture is an X-ray.

Digital X-ray machines are accurate in detecting fractures in the extremities and following significant injury. However, X-rays are known to miss fractures observed in hips, ribs, spine, and wrist if not evident enough.

Following the X-ray, a radiologist would have to screen through the X-ray to conclusively diagnose a fracture. Several factors contribute to how accurately a radiologist diagnoses an X-ray. Experience, the number of hours worked, multitasking, errors based on speed, and even the physical environment around the radiologist can all contribute to a missed diagnosis of a fracture.

Additionally, technical errors such as alterations in the quality of images or obtaining an X-ray in the incorrect view can also contribute to a misdiagnosis of a fracture.

Some fractures are not displaced sufficiently to be picked up on an X-ray, and subsequently by a radiologist.

How can AI be used in radiology?

AI algorithms are taking significant steps to negate some of the technical and human factors that play into fracture diagnosis.

Deep learning, a subset of machine learning, is working to identify features of radiological imaging. This is primarily due to the big data that is available to train systems along with the fine-tuning of algorithms to pick up even the smallest changes noted on radiographs.

An example of a deep learning model used in modern imaging is the convolutional neural network (CNN). Results from a CNN system are output only after significant reasoning implemented by the multi-layered system. All this usually takes only a few minutes to compute.

The benefits of such systems are reducing time to diagnosis and eliminating errors along the way. This increases the volume of data that can be computed in a day with increasing accuracy.

Innovations such as deep learning technology are minimizing the learning curve required for radiologists to implement AI tech into medical imaging.

Improved accuracy in fracture diagnosis

Misinterpretation of fractures on X-rays can be as high as 24% in emergency room settings.

AI is being used to improve both the accuracy and efficiency of fracture diagnosis, and AI models can be made to run both supervised and unsupervised.

For radiological imaging, supervised machine learning is encouraged, where data set outcomes have already been predicted. This helps to train systems to assess for fractures among particular X-rays. The system also continues to iterate itself over time to predict fracture diagnosis more accurately.

The key outcome is to reduce errors in diagnosis and speed up management for patients who do present with fractures.

Additionally, AI models can be used to diagnose fractures in locations such as ribs, hips, and vertebrae. Due to the possibility of minute and multiple fractures in these locations, especially among the elderly, errors in diagnosis using an X-ray are large.

Providing CNNs with large volumes of data in this regard helps to facilitate models that can reduce such errors, improving diagnostic accuracy.

AI models can also be used to highlight regions of interest within digital X-rays. These can then be used to understand whether fractures were present or not. If present, the data can be used to improve readings for similar findings in the future.

AI is the future of fracture diagnosis

AI-assisted systems prioritize X-rays where fractures are predicted to be present. Within emergency room settings, this can reduce patient wait times in cases where fractures are not present.

AI can facilitate preliminary results, guiding radiologists toward diagnostic possibilities. AI systems can also help diagnose serious cases with seemingly subtle symptoms.

Specialists can review results for less severe cases when time permits.

It will take time to refine these systems to provide a near-100% accuracy. It will also be crucial to factor in cases where false-positive or false-negative results are noted. Machines are also free from liabilities when it comes to decision-making. This means that in the near future radiologists will continue to be the sole interpreters of X-rays, even if it is with the efficient assistance of AI-generated systems.


Dr. Michelle Frank is a healthcare consultant working on building safe online health communities for women. She works with startups building digital products enabling women to make informed decisions for their health. She resides in India, exploring different cities and opportunities as they arise. Feel free to join in for #MyHealthChat which she co-hosts every alternate Thursday on Twitter.

Main image

Photo by MART PRODUCTION from Pexels