Machine learning applied to gait performance in patients with type 2 diabetes.
Diabetes Mellitus; Gait; Machine Learning
Introduction: Diabetes is characterized by a metabolic disease group that
can cause several changes. One of them occurs in the sensorimotor function,
which generates alterations in gait execution, such as longer stance phase,
shorter steps and inadequate plantar pressure distribution. Quantitative
methods for assessing changes in gait patterns can be decisive in designing
treatment strategies. In addition, they can help prevent diabetes
complications. With advances in machine learning (ML) techniques,
automated pattern recognition in the face of large amounts of data has
become an essential tool in the medical field due to the ability to predict
clinical complications before the disease gets worse. Objectives: to apply ML
models on gait evaluation data of type 2 diabetic patients to identify patterns
of gait performance that may predict diabetes clinical complications.
Methods: The study involved two methodological modalities: 1) Protocol
elaboration and Systematic Review; 2) Development and improvement of
predictive models of unsupervised and supervised BF for exploratory data
analysis, diabetes detection and detection of clinical complications in diabetes
based on glycated hemoglobin levels. The data for the execution of the study
was provided through a partnership with the International University of Florida
(FIU) during a sandwich doctorate (Edital No. 02/2020 – CAPES/PRINT)
between September 2021 and June 2022. The data were pre-processed and
implemented in different ML models. The ML models used efficiency were
evaluated based on the silhouette analysis for unsupervised ML, the
confusion matrix for supervised ML metrics, and conventional statistics,
adopting a significance level of 5%. Results: 1) Resulted in two articles:
Article 1 - The protocol defined the systematic review methodology; Article 2 -
The review included four studies (208 participants). Two used ML as a
predictive method, one used conventional statistics based on multiple
stepwise regression, and the last used the uncertainty method Fuzzy
classifier. Studies achieved at least 75% in adequately reporting 19 TRIPOD
items. Three of the included studies were classified as high risk of bias. 2)
Resulted in three articles: Article 3 – K-Means separated the dataset into two
groups (silhouette = 0.47). Gait speed, step length and plantar pressure
distribution patterns were statistically different (p < 0.05) between diabetics
and non-diabetics. Furthermore, among diabetic patients, a statistically
significant difference (p < 0.05) was observed in plantar pressure distribution
patterns. Article 4 – Supervised ML algorithms using gait data showed high
sensitivity in the distribution of plantar pressure in the heel region to classify
diabetes from non-diabetics. Article 5 – The XGB classifier presented better
results, reaching an AUC of 0.99, a precision of 0.91, a recall of 0.90 and an
f1-score of 0.89. The three most relevant gait characteristics in the
classification of diabetes complications found were left support base, mean
left pressure over time in the metatarsal region (I-III) and mean active sensor
area in the III-IV phalanges region. Conclusion: The use of ML has been
growing more and more, making it necessary to have more rigorous
methodological studies to improve the evidence regarding the development
and application of predictive tools, mainly concerning the gait performance
data of diabetic patients. ML in medical applications has been helping to
develop new skills and approaches in the way of treating. Regarding diabetic
patients, the use of ML in the early detection of alterations in gait performance
may be fundamental in clinical practice for directing treatments, which may be
crucial in preventing the appearance of neurological disorders.