Validation of a Semi-Automatic Method for Computing Distal Radius Radiographic Measurements
2Orthopaedics, Hadassah Medical Center, Israel
3School of Computer Science and Engineering, the Hebrew University of Jerusalem, Computer-Assisted Surgery and Medical Image Processing Laboratory, Israel
Background:
The most commonly used surgical criteria for decision making besides age, level of activity and hand dominance in distal radius fractures are based on radiographic parameters. These measurements which aid clinical decision making are not routinely used by all caregivers as they are time consuming. The goal of our study was to aid the treating physician in the decision-making process. We aimed to produce an automatic fracture measurement tool. The program computes selected measurements and presents them to the treating physician together with the radiographic images. In this study we show the validation process of this semi-automatic method.
Methods:
The research was conducted in collaboration between the Hadassah Medical Centre and the School of Computer Science and Engineering, The Hebrew University of Jerusalem. An algorithm was developed to automatically compute specific radiographic measurements from anatomical landmark markings in the radiographs. In the anteroposterior (AP) radiograph, articular step and gap, radial inclination, radial length, radial shift and ulnar variance were computed In the lateral radiograph articular gap and step, palmar tilt and dorsal shift were computed. The validation process consisted of the following steps. A total 22 AP and lateral radiographs of patients with distal radius fracture from the Hadassah data base were collected. Using a computer software (ITK-SNAP 3.8 www.itksnap.org) these cases were segmented and the above measurements were calculated by an orthopaedic trauma fellow and a highly experienced fellowship trained orthopaedic trauma surgeon. These measurements were compared to quantify the inter and intra observer reliability. After the development of the algorithm, the automatically segmented fracture parameters were compared to the manual measurements. Automated calculated measurements that are within the interobserver variability are considered accurate in this study. The computation of overall accuracy was done by assigning all measurement classes an equal weighting and taking an average of the accuracy.
Results:
In the anteroposterior algorithm we obtained 92.7% accuracy. In the automated results, the radial angle and ulnar variance have one measurement out of the observer variability range, with a 94.7% accuracy each. All measurements of radial length and radial shift are 100% accurate. Gap measurements were 66% accurate, and step measurements were 100% accurate. In the lateral algorithm we obtained 100% accuracy. Palmer tilt, dorsal shift, gap and step measurements were 100% accurate.
Conclusion:
This semi-automated method for computing radiographic measurements of distal radius fracture is highly accurate. Further steps of validation and development of a fully automated method are currently under development.