Artificial Intelligence Makes Early Arthritis Detection Possible

Date:9 June 2022 Author: Juandre

Artificial intelligence neural networks have been trained to distinguish between two types of arthritis and healthy joints, according to researchers.

The neural network correctly identified 82 percent of healthy joints and 75 percent of rheumatoid arthritis cases. When paired with a doctor’s experience, it has the potential to produce far more accurate diagnoses. This approach will be investigated further in a future initiative, according to the researchers.

Because there are so many distinct types of arthritis, pinpointing which one is hurting a patient’s joints can be challenging. In an interdisciplinary research effort, computer scientists and clinicians from Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) and Universitätsklinikum Erlangen taught artificial neural networks to discriminate between rheumatoid arthritis, psoriatic arthritis, and healthy joints.

A team led by Prof. Andreas Maier and Lukas Folle from the Chair of Computer Science 5 (Pattern Recognition) and PD Dr. Arnd Kleyer and Prof. Dr. Georg Schett from the Department of Medicine 3 at Universitätsklinikum Erlangen was tasked with investigating the following questions as part of the BMBF-funded project “Molecular characterization of arthritis remission (MASCARA).” Is it possible for artificial intelligence (AI) to distinguish between different types of arthritis based on joint shape patterns? Is this method effective for making more exact undifferentiated arthritis diagnoses? Is there any portion of the joint that has to be examined more closely during the diagnosis?

Due to a lack of biomarkers, correctly categorizing the appropriate kind of arthritis is currently difficult. X-ray images used to aid diagnosis are also unreliable due to their two-dimensionality, which is insufficiently exact and allows for interpretation. This is in addition to the difficulty of positioning the joint for X-ray imaging.

Finger joints are used to teach artificial networks

The researchers concentrated their observations on the metacarpophalangeal joints of the fingers, which are frequently afflicted early in patients with inflammatory illnesses such rheumatoid arthritis or psoriatic arthritis. With the goal of distinguishing between “healthy” joints and those of patients with rheumatoid or psoriatic arthritis, a network of artificial neurons was trained using finger images from high-resolution peripheral quantitative computer tomography (HR-pQCT).

HR-pQCT was chosen because it is currently the best quantitative method for obtaining high-resolution three-dimensional images of human bones. Changes in the structure of bones can be observed extremely precisely in the case of arthritis, making exact classification possible.

More tailored treatment may be achievable using neural networks

The artificial network was then tested on a total of 932 fresh HR-pQCT images from 611 patients to see if it could truly implement what it had learned: Is it capable of accurately assessing the previously classified finger joints?

The results indicated that AI diagnosed 82 percent of healthy joints, 75 percent of rheumatoid arthritis cases, and 68 percent of psoriatic arthritis cases without any further information, which is a very high hit probability. It could lead to considerably more accurate diagnoses when combined with the expertise of a rheumatologist. The network was also able to appropriately classify cases of undifferentiated arthritis when presented with them.

Latest Issue :

May-June 2022