Skip to content

Artificial Intelligence Improves Orthopedic Diagnosis

In This Article

  • Researchers at the Foot & Ankle Research and Innovation Laboratory (FARIL) at Massachusetts General Hospital have created an innovative automated musculoskeletal image interpretation system (AMIIS) that improves diagnostic accuracy to more than 90%
  • AMIIS harnesses artificial intelligence (AI) to interpret patients' data and clinical images, reducing misdiagnoses caused by lack of experience, expertise, time and access to modern imaging techniques
  • AMIIS is part of a larger FARIL project to improve musculoskeletal imaging modalities, educate providers in all orthopedic settings about emerging imaging techniques and diagnostic methods and democratize knowledge and expertise globally.

Investigators in Massachusetts General Hospital's Foot & Ankle Research and Innovation Laboratory (FARIL) in the Department of Orthopaedic Surgery have developed a novel automated musculoskeletal image interpretation system (AMIIS) to improve diagnostic accuracy in patients with musculoskeletal (MSK) conditions. The development of artificial intelligence (AI)-based systems such as AMIIS is a key aspect of FARIL's project studying the application of AI in the diagnosis of orthopedic conditions.

"Improving current assessment methods and imaging modalities will enhance diagnostic accuracy in these patients," says Soheil Ashkani, MD, a physician investigator/fellow who is the project's leader. "Our studies' outcomes have shown that the application of AI improves diagnostic accuracy to more than 90%. Using this AI-based system will significantly reduce the number of missed cases by clinicians and health care providers who sometimes miss occult fractures and subtle joint instabilities on primary X-rays, particularly in the foot and ankle, in up to 50% of patients.

The FARIL team includes AI-engineer Reza Mojahed-Yazdi, MSc, Bart Lubberts, MD, PhD, R&D Director, Christopher DiGiovanni, MD; Chief of the Foot and Ankle Center, Foot and Ankle Surgeons Daniel Guss, MD, MBA, and Gregory Waryasz, MD; Foot and Ankle Research Fellow Rohan Bhimani, MD, MBA; and Dr. Ashkani.

Challenges in Musculoskeletal Imaging

Dr. Ashkani says that early, accurate diagnosis of orthopedic patients is sometimes even more important than treatment. Faster, more precise diagnoses lead to significant morbidity and mortality reduction and lower costs for the health care system as well as the patients. But detection of MSK conditions relies highly on clinical imaging techniques, such as X-rays, CT scans, MRIs and ultrasounds, as well as patient history, physical examination and lab assessments.

"Clinicians and emergency care providers may sometimes miss injuries, especially when they are subtle or occult, and particularly when there is a lack of experience, expertise, time or access to modern imaging techniques," says Dr. Ashkani.

With respect to foot and ankle, occult and subtle injuries such as occult ankle fractures, metatarsal fractures and subtle instabilities of the syndesmosis and Lisfranc joint illustrate this difficulty because clinicians do not always appreciate the injury on the primary imaging and examinations.

"Sometimes they cannot simply say that there is a fracture in the bone, especially when occult, or if a specific amount of diastasis indicates an unstable syndesmosis," says Dr. Ashkani. "That's why we need to dig deep into these clinical images and other patient’s data. Together with our lead AI-engineer, Reza Mojahed-Yazdi, we developed an algorithm that can receive a huge amount of patients’ data, classify them, and use them to determine abnormalities. Other than conventional X-rays and CT scans, we also incorporate weight-bearing imaging as an important parameter in lower extremity assessment, using weight-bearing CT scan and weight-bearing X-rays. Imaging in weight-bearing conditions leads to better and more accurate visualization of the bones and joint’s congruence and helps clinicians to see abnormalities better. In non-weight-bearing condition, detection of joint abnormalities and congruence becomes harder, especially if subtle."

Harnessing AI in Radiology to Diagnose Orthopedic Conditions

Recent advances have led investigators to widely apply AI to the interpretation of patients' data and clinical images. As a result, there has been a noticeable achievement in the prediction and detection of various clinical conditions, especially in orthopedic surgery.

In previous reports on ankle fractures, it was shown that in limited-resource settings, up to 50% of the cases might be missed on initial X-rays. "In earlier research, we showed that using weight-bearing CT scans gave a diagnostic accuracy of about 79% in cases with subtle syndesmotic instability or 82% in subtle Lisfranc instability," says Dr. Ashkani. "However, in our most recent research using AI for detection of ankle fractures as well as subtle syndesmosis and Lisfranc instabilities on both weight-bearing and non-weigh-bearing CT scan and X-rays, we achieved accuracies of more than 90%. As our next step to expand our work, we have put our focus on using AI-based interpretation methods on X-rays as a primary imaging modality that is available in almost every clinical setting. We will then incorporate conventional and weight-bearing CT scans, aiming to improve the accuracy of interpretation while reducing the costs and time consumed by the health care providers and the patient. With the help of AMIIS, the amount of missed cases by health care providers, residents and those who might not be experts in finding these illnesses will significantly reduce, and consequently, the rate of long term disabilities and comorbidities will decrease."

"Previous reports used a smaller population in their studies and gave diagnostic accuracy of 82%," says Dr. Ashkani. "In our collaboration with other labs, we achieved 90% using CT scans and AI. Now, we are using X-rays, which are available in every clinical setting, and getting more than 90%—sometimes up to 99.8%—for Lisfranc ankle fractures. Without the help of AMIIS, residents and health care providers who are not experts in finding these fractures will miss up to 20–30%. We will not miss even one using this system."

In the project's first phase, they entered weight-bearing and non-weight-bearing X-rays and conventional and weight-bearing CT scans—the most useful imaging modalities in orthopedic settings—into AMIIS in addition to other data from the patient. Modern algorithms for developing deep neural networks, natural language processing and image analysis techniques were used to process the data.

Dr. Ashkani notes, "AMIIS examines images pixel by pixel. So even if you have an occult fracture not visible and easy to miss on X-rays, or hidden behind bone complexes, AMIIS will highlight the point of abnormality and flag that so you can identify the orthopedic condition."

FARIL's AI Project Goals and Collaborations

The Application of AI in the Diagnosis of Orthopedic Conditions project has three goals:

1. Improving current modalities

FARIL investigators are seeking to improve the image and patient data interpretation in emergency departments, orthopedics, rheumatology, podiatry and trauma settings, especially those with less advanced imaging devices and limited expertise. "We have created the AI platform for these conditions. We now need to collaborate and validate our system in other centers," Dr. Ashkani says. "We have prioritized these patients based on the importance and severity of their MSK conditions and we are collaborating with other divisions to include common injuries in other parts of the MSK system as well."

The FARIL team's AI project involves several key collaborations. FARIL is working with the Harris Orthopaedics LabOrhun Muratoglu, PhD, lab director, and his team are working closely with Dr. Ashkani and his colleagues on applying machine learning algorithms to weight-bearing computed tomography (CT) images.

2. Educating providers in emerging imaging modalities

Their second aim involves educating clinicians and health care providers in the use of modern imaging devices such as portable ultrasound and weight-bearing CT scans and weight-bearing MRI in the future. Moreover, FARIL aims to provide educational content for health care providers about different aspects of orthopedic conditions, from the anatomy and pathophysiology to diagnosis and treatment.

"Portable ultrasound is a newly emerging, cheaper imaging technique that can easily be used on the sports field, battlefield or at the site of an accident or trauma. Clinicians connect the portable ultrasound to a cell phone or a tablet, allowing them to conduct on-site examinations without monitors or other equipment. Portable ultrasound can help clinicians in making better care decisions and interpret images more precisely, even outside of the clinic," adds Dr. Ashkani.

"AMIIS can help clinicians make better care decisions and interpret images more precisely, even outside of the clinic," adds Dr. Ashkani. "We are providing educational content, such as videos and slideshows, for clinicians to learn to use portable ultrasound to detect conditions like syndesmotic instability or ankle fractures even without X-rays."

In another partnership with the Harvard Global Orthopaedics Collaborative (HGOC), Dr. Ashkani's team is making educational videos for the most common orthopedic conditions. So far, they have created three sample videos on ankle fractures:

Currently, these videos are available on HGOC's YouTube channel and FARIL's website for use by orthopedic care providers in low and middle-income countries. "Educating people globally with the new diagnostic methods and interpretation techniques will lead to spreading the knowledge and experience all around the world," notes Dr. Ashkani. This team is also gathering feedback from Mass General residents, clinicians and surgeons, as well as providers in other settings, particularly those with limited resources and expertise, to create a larger and more comprehensive library in this regard.

3. Increasing AI learning

Finally, FARIL wants to provide a platform for clinicians to help improve the performance of AMIIS and the quality of the educational content.

"The more opinions, feedback and data we get, the better performance we have. AMIIS receives more data every time a clinician uses it, so it will also evolve to be smarter and encompass a bigger data source," says Dr. Ashkani. "The ultimate plan is to make a comprehensive AI-based product that predicts orthopedic conditions with high accuracy and helps the user manage the patient properly. This product will not only improve diagnosis precision but also can be a means for educating and updating health care providers around the world about new methods of diagnosis and treatment in orthopedic surgery. This is a great step forward in democratizing knowledge, experience and expertise globally."

Expanding AMIIS's Reach

After securing a patent for AMIIS, turning it into intellectual property, and creating the prototype, Dr. Ashkani and his team are seeking further support and collaboration with industry and other funding resources to expand the product. The FARIL team aims to create a user-friendly easy-access product that makes the clinicians able to assess and manage the patient quickly, precisely and sufficiently.

"When you are in a think-tank with world-class scientists, engineers and clinicians, and you have the privilege to work on these kinds of projects here at Mass General, what else do you really need?" says Dr. Ashkani.

Learn more about the Foot & Ankle Research and Innovation Lab

Refer a patient to the Department of Orthopaedic Surgery


The Foot & Ankle Research and Innovation Laboratory (FARIL) at Massachusetts General Hospital harnesses next-generation imaging techniques to more completely assess orthopaedic foot and ankle pathology.


Although not definitive, a systematic review showed support for treating acute tuberosity (zone 1) avulsion fractures of the 5th metatarsal conservatively, and treating all acute Jones (zone 2) fractures of the 5th metatarsal surgically.