AI detects hip malpositions

AI

Hip deformities in newborns can require many operations. SCS has developed an AI that detects such deformities.

  • Problem

    A hip deformity in newborns can make various operations necessary in the course of life. To prevent this, it is sufficient to determine the Graf alpha and beta angles on the basis of an ultrasound image of the hip and, if necessary, to put on a brace for a few months for treatment. Doctors' current diagnoses are subjective and prone to error.

  • Solution

    Using training data consisting of ultrasound images, SCS was able to train various algorithms that allow an objective determination of the angles. In this way, the doctor can be supported.

  • Added value

    Doctors are alerted at an early stage to an error in the image acquisition and can rely on an objective and accurate angle determination. Many hip operations in the later future can thus be prevented.

Project insights

About 3 percent of all newborns are born with a hip deformity. If this deformity is detected by ultrasound examinations, it can be easily treated by putting a corset on the baby for several weeks (Figure 1). Undetected, this so-called hip dysplasia leads to hip surgery before the age of sixty in 30% of those affected.

The criterion for whether a malposition develops is the alpha angle according to Graf's method. An evaluation of the angles determined by doctors shows a strong tendency towards "borderline cases" at 60degrees that do not require treatment.

Picture left: Infant with flexion orthosis. Picture right: Skeleton model with drawn-in image plane of the ultrasound images.
Picture left: Infant with flexion orthosis. Picture right: Skeleton model with drawn-in image plane of the ultrasound images.

As an alternative, Machine Learning and Data Analytics - especially thanks to developments in image processing using Deep Learning - offers a way to determine the angle objectively and reproducibly. To confirm this expectation, SCS has conducted a study together with various experts and paediatricians [1] that can confirm this thesis [2].

Picture left: Distribution of angles determined by doctors in the dataset of the study. Picture on the right: Ultrasound image with lines drawn in for angle determination (yellow: doctor, red: algorithm).
Picture left: Distribution of angles determined by doctors in the dataset of the study. Picture on the right: Ultrasound image with lines drawn in for angle determination (yellow: doctor, red: algorithm).

For this purpose, the data basis was first cleaned by removing the overhang of borderline cases. Subsequently, a deep neural network was trained that predicts the lines on ultrasound images that determine the hip angle, analogous to the medical procedure. A comparison with data from paediatricians that were not used for training shows that our algorithm can reliably determine the angle.

[1] Partners: Stefan Essig MD, Thomas Baumann MD

[2] Reference: Oelen D, Kaiser P, Baumann T, et al. Accuracy of Trained Physicians is Inferior to Deep Learning-Based Algorithm for Determining Angles in Ultrasound of the Newborn Hip Ultrasound Med. 2020;10.1055/a-1177-0480. doi:10.1055/a-1177-0480.

Related projects

IOBnet - Exchange platform for ophthalmologists

The IOBnet platform offers ophthalmologists the possibility to exchange anonymized cases with experts at any time. With the creation ... learn more

Cloud platform for medical measurement data

The spirometry devices from ndd Medizintechnik AG make a significant contribution to the early detection of lung diseases. Their measurement data are always ... learn more

Predictive Alarm Detection

SBB ZKE measures the bearing and wheel temperatures of passing trains on the open track. SCS has trained an AI model that can measure the ... learn more

Detection of wheel faults with smart algorithms

Wheel faults lead to high stress on the railway infrastructure and generate noise. SCS has developed algorithms to detect wheel faults. Thus ... learn more

AI at the edge - counting cells

SCS developed a deep learning algorithm that determines the number of cells in an image in the microscopy system. The AI was developed for ... learn more
Show all projects
Dr Christof Bühler Life Science & Physics How can I help you?