APPLICATION OF ARTIFICIAL INTELLIGENCE METHODS IN FLATFOOT ASSESSMENT
Birutė Sinkutė
St. Ignatius Loyola College, Kaunas, Lithuania
Justina Šeštokė
St. Ignatius Loyola College, Kaunas, Lithuania, Prof. K. Barsauskas Ultrasound Research Institute, Kaunas University of Technology, Lithuania
Eglė Butkevičiūtė
Department of Software Engineering, Faculty of Informatics, Kaunas University of Technology, Lithuania
Keywords: orthopaedics; flat feet; artificial intelligence; 2D images.
Abstract:
This study examined the potential of artificial intelligence tools for detecting flatfoot pathology. We want to emphasise that there is very little research in this area and to point out that this is a relevant and very important topic in medicine. First, the base flow used a pretrained “backbone” on the ImageNet dataset. In this study, this term refers to the feature extraction part of a convolutional network. A standardised pre-processing with pruning and augmentation was performed, and a three-stage training schedule (stages 1, 2, and 3), average and maximum aggregation at the subject level, and the addition of light test time were proposed. Nine different model architectures were used. From stage 2 onwards, all models were trained on feet. Three-dimensional photographs with real flatfoot shapes, from flatfoot stages I to IV, were used. The most validated model was displayed in accurate AUROC plots, with estimated average and maximum aggregation values and their standard deviations. The research and calculations demonstrate the feasibility of applying artificial intelligence in orthopaedics. This work aimed to apply artificial intelligence methods and to detect flat feet.
DOI: