Can Artificial Intelligence solve one of the big problems in shoulder arthroplasty research?.

We now recognize that most of our questions about shoulder arthroplasty cannot be answered using a few hundred shoulders. We also know that plain anteroposterior and axillary radiographs (not CT scans) are the most practical images for evaluating the geometry of shoulder arthroplasty.

 If we want to find out what geometric parameters determine the clinical result of anatomic or reverse total shoulder arthroplasty, we need BIG data, i.e. an assessment of tens of thousands of preoperative and postoperative plain radiographs to identify the characteristics that affect the patient's outcome. 

If performed manually by a human,  this labor would be prohibitive in time, cost and effort, even for Hercules. However, the task is well suited to artificial intelligence by which algorithms can be trained to rapidly categorize and - eventually - to analyze very large numbers of anteroposterior and axillary plain x-rays before and after shoulder arthroplasty so that the geometry of the reconstruction can be correlated with the clinical outcome.

As a step in this direction, the authors of Deep learning to automatically classify very large sets of preoperative and postoperative shoulder arthroplasty radiographs developed an artificial intelligence (AI) algorithm to classify preoperative and postoperative radiographs from shoulder arthroplasty patients according to laterality, radiographic projection, and implant type.

They used 2303 plain x-rays from 1724 shoulder arthroplasty patients. Two observers manually labeled all radiographs according to (1) laterality (left or right), (2) projection (anteroposterior, axillary, or lateral), and (3) whether the radiograph was a preoperative radiograph or showed an anatomic total shoulder arthroplasty or a reverse shoulder arthroplasty. The 18 classes of x-rays are shown below.

The labeled plain x-rays were randomly split into developmental and testing sets. 

A deep-learning algorithm was trained on the developmental set to classify the 3 characteristics described above. 

The trained algorithm was then evaluated on the testing set.

The trained algorithm perfectly classified laterality (right vs left).

The algorithm achieved accuracy scores of 99.2%, 100%, and 100% in identifying anteroposterior, axillary, and Y views

The algorithm achieved accuracy scores of 100%, 95.2%, and 100% in identifying preoperative radiographs, anatomic total shoulder arthroplasty radiographs, and reverse shoulder arthroplasty radiographs 

It took the algorithm 20.3 seconds to analyze 502 images.

The model focused on certain relevant pixel regions. For example, for the classification of prosthesis type, the model selected the pixels around the edge of the implant as shown by the green dots shown below.

From the examples provided, it appears that the imaging protocols for the different x-ray views were well standardized; i.e. all of the axillary views show the same angle of the beam and position of the arm. It would be of interest to know whether there needs to be a step that excludes non-standard projections such as that shown below.

Comment: For any of our readers it would be straightforward and quick to classify one shoulder radiograph regarding laterality, view, preoperative vs. postoperative, and anatomic vs. reverse arthroplasty. However, classifying and recording this information for thousands of radiographs would become tedious and time-consuming. The ability to complete these tasks automatically, quickly and precisely is an important step in the use of artificial intelligence to classify shoulder x-rays. 

You can support cutting edge shoulder research that is leading to better care for patients with shoulder problems, click on this link

Follow on twitter/X: https://x.com/RickMatsen
Follow on facebook: https://www.facebook.com/shoulder.arthritis
Follow on LinkedIn: https://www.linkedin.com/in/rick-matsen-88b1a8133/


Here are some videos that are of shoulder interest
Shoulder arthritis - what you need to know (see this link).
How to x-ray the shoulder (see this link).
The ream and run procedure (see this link).
The total shoulder arthroplasty (see this link).
The cuff tear arthropathy arthroplasty (see this link).
The reverse total shoulder arthroplasty (see this link).
The smooth and move procedure for irreparable rotator cuff tears (see this link).
Shoulder rehabilitation exercises (see this link).