Prediction of the facial growth direction with Machine Learning methods

Abstract

First attempts of prediction of the facial growth (FG) direction were made over half of a century ago. Despite numerous attempts and elapsed time, a satisfactory method has not been established yet and the problem still poses a challenge for medical experts. To our knowledge, this paper is the first Machine Learning approach to the prediction of FG direction. Conducted data analysis reveals the inherent complexity of the problem and explains the reasons of difficulty in FG direction prediction based on 2D X-ray images. To perform growth forecasting, we employ a wide range of algorithms, from logistic regression, through tree ensembles to neural networks and consider three, slightly different, problem formulations. The resulting classification accuracy varies between 71% and 75%.

Prediction of the Facial Growth Direction is Challenging

“…In our other article, we described in detail the attempts to date to predict craniofacial growth [12]. In summary, none of the methods to date have been sufficiently effective when validated on a sample other than the one used to develop the method.…”

“…In our previous study, we described the datasets in detail [12]. In short, subjects from "Nittedal Growth Study" housed in the University of Oslo, Norway [8] and American Association of Orthodontists Foundation (AAOF) Craniofacial Growth Legacy Collection (Bolton-Brush, Burlington, Denver, Fels Longitudinal, Forsyth Twin, Iowa, Mathews, Michigan, and Oregon collections) [1] were included in this investigation.…”

“…Further difficulties are related to the low number of dataset instances, which makes any deep architecture prone to overfitting and class imbalance. Last but not least, it is worth pointing out that we are restricted to work with 2D LC's due to the ethical issues described in [12].…”

“…The third set of features, transformed coordinates, is created by moving all raw landmarks so that the Sella point is positioned at (0, 0). Justification of such transformation and its visualization are presented in [12]. The total number of cephalometric features from one timestamp amounts to 15, Procrustes -40, and transformed -40.…”

“…• addressing the problem of FG prediction with ML methods, which, to our knowledge, was analyzed only once before [12] and improving the prediction score by 2.81%;…”