194 - Unique vibroacoustic biosignatures demonstrated in two animal species; a possible leap in animal models of human disease.

Poster Number: 194
Publication Number: 194.2273

Background: Living tissue within body systems, such as the circulatory, respiratory, and digestive systems, emit sounds and vibrations as a function of performing biological processes. These sounds can fall within the infrasonic-to-ultrasonic frequency spectrum. Current sensor technologies fail to capture all inaudible sounds and vibrations or discard them as noise. These overlooked or discarded data may contain key health information.

Objective: In this study, we used a novel infrasound-to-ultrasound e-stethoscope (imPulseTM UNA – Fig. 1) and artificial intelligence to examine if captured audible and inaudible sounds and vibrations from two animal species could be discerned.

Design/Methods: We trained and tested a machine learning classification algorithm. Vibroacoustic recordings were collected at eight different parts of the body (Fig. 2) from 18 cats and 39 dogs of varying breeds, sizes, ages, and sex. Recordings from these eight auscultation points, individually and in three unique combinations, were used for analysis. Recordings were segmented into non-overlapping 10-second intervals and were used to train a machine learning model to make species classification predictions. The model utilized transfer learning from pretrained deep neural network audio classification models (e.g., YAMNet). Accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC) were used to measure model performance.

Results: Our model discriminated between species with 93% accuracy, 86% sensitivity, 97% specificity, and AUC of 0.97 (Table 1). We anticipate improved class prediction as our feline and canine database continues to grow.

Conclusion(s): We demonstrate that pretrained audio classification algorithms successfully discriminate between cats and dogs from audible and inaudible sounds obtained from the right GI tract with >90% accuracy. It is intriguing why the right gastrointestinal (GI) tract generates biosignatures that can discern between species and whether the microbiome or other factors in these two species are responsible. Our results suggest that such vibroacoustic biosignatures can be extended to distinguish between physiological and pathological states. When examined in conjunction with human data, evaluation of such traits makes the animal models more relevant and may shed light into overall biological changes that are difficult to differentiate using current imaging, ultrasound, or molecular techniques. Further research is ongoing in humans and animals to determine the basis of such unique biosignatures, which may improve medical prediction, diagnosis, and prevention.