Acoustic Frequency Analysis Investigating Frequency Differences in Acute Chest Syndrome

Acute chest syndrome (ACS) is the leading cause of death among people with sickle cell disease, which affects the hemoglobin in red blood cells causing them to sickle in shape. ACS is clinically defined and diagnosed by the presence of a new pulmonary infiltrate on chest imaging with accompanying fever and respiratory symptoms like hypoxia, tachypnea, and shortness of breath. However, the characteristic chest x-ray (CXR) findings necessary for a clinical diagnosis of ACS can be at times difficult to detect, as is determining which patient needs a CXR. This makes early detection hard, but it is critical for limiting the severity of ACS and subsequent fatalities. The other diagnostic method currently used is computed tomography (CT), but these are more expensive and expose patients to more radiation. This research project looks to apply percussion and auscultation techniques that can provide an immediate diagnosis of acute pulmonary conditions by using a standardized input called the Tabla and electronic auscultation for post frequency analysis. Sickle cell patients having acute chest syndrome, vaso-occlusive crisis (VOC), and regular clinic visits were recorded from and analyzed using MATLAB. Those subjects with a clear clinical diagnosis of ACS were analyzed separately to compute the frequency spectrum of the recordings. These spectrums were averaged to give the ACS subject pool average and the same was done for the healthy subject group. This analysis revealed that there was an average of 10-14 dB decrease in sound intensity in the ACS subject group compared to the healthy group. Because of acute chest syndrome’s complex clinical presentation and pathogenesis, a thorough retrospective analysis of each patient’s EMR was performed to identify confounding variables that might remove them from the subject pool. This review along with frequency analysis revealed a new classification label for subjects who had complicated VOC, meaning some sort of pulmonary infiltrate was present. A random under-sampling boosted tree classification model was performed identifying with 89% accuracy the positive ACS observations and healthy observations. Future work will continue to gather more ACS data to continue to train and test more classification algorithms.