Machine learning for automatic identification of thoracoabdominal asynchrony in children

Automated Assessment of Thoracic-Abdominal Asynchrony in Patients with Morquio Syndrome

“…The model was trained using an elastic-net regularization algorithm and with signals recorded from healthy pediatric volunteers while they were breathing normally and with a resistive load to simulate abnormal breathing. It was shown that the model assessed pulmonary dysfunction among children with neuromuscular disorders very successfully and was consistent with expert pulmonologists’ evaluations [ 15 ]. Because people with Morquio syndrome have small stature with lung capacities similar to those of children, the same ML model to evaluate respiratory function among subjects with Morquio syndrome was used.…”

“…The lag between thoracic and abdominal movements is known as thoracic-abdominal asynchrony (TAA), indicating the severity of pulmonary dysfunction. Other parameters such as percentage RC (%RC), which is an indication of the RC contribution to Vt, and the labored breathing index (LBI), which is a measure of additional respiratory effort due to asynchronous breathing, are other indices computed to assess TAA [ 14 , 15 ].…”

“…Accurate interpretation of the breathing patterns, by observing their variability over time and without outlier bias, is crucial for predicting impending respiratory dysfunction and failure. Hence, a median-based robust statistic that uses a cumulative frequency metric, specifically the inverse cumulative percentage (ICP), was developed to assess variation over time [ 15 ]. A novel visual representation of ICP identifies the median statistic and provides relative durations of normal and abnormal breathing in the patient.…”

“…A novel visual representation of ICP identifies the median statistic and provides relative durations of normal and abnormal breathing in the patient. Pulmonologists found the graphic display of ICP very useful, intuitive, and easy to interpret [ 15 ].…”

“…Further, using the ICP parameter, an accurate automated machine learning (ML) system for the classification of normal and abnormal breathing was developed and published [ 15 ]. The model was trained using an elastic-net regularization algorithm and with signals recorded from healthy pediatric volunteers while they were breathing normally and with a resistive load to simulate abnormal breathing.…”