[Paper Review] Instant Preliminary Cardiac Analysis from Smartphone Auscultation: A Real-World Canine Heart Sound Dataset and Evaluation

Aswin Jose, Roeland P. J. E. Decorte+1 more|arXiv (Cornell University)|Jan 20, 2026

Phonocardiography and Auscultation Techniques0 citations

TL;DR

The paper presents a real-world canine heart sound dataset collected from smartphones and evaluates SoNUS 3.2.x for preliminary cardiac analysis, focusing on heart rate estimation under variable recording conditions and introducing a quality scoring module and a fast 30–40s variant.

ABSTRACT

This study presents a real-world canine heart sound dataset and evaluates SoNUS version 3.2.x, a machine learning algorithm for preliminary cardiac analysis using smartphone microphone recordings. More than one hundred recordings were collected from dogs across four continents, with thirty eight recordings annotated by board certified veterinary cardiologists for quantitative evaluation. SoNUS version 3.2.x employs a multi-stage fallback architecture with quality-aware filtering to ensure reliable output under variable recording conditions. The primary sixty second model achieved mean and median heart rate accuracies of ninety one point six three percent and ninety four point nine five percent, while a fast model optimized for thirty to forty second recordings achieved mean and median accuracies of eighty eight point eight six percent and ninety two point nine eight percent. These results demonstrate the feasibility of extracting clinically relevant cardiac information from opportunistic smartphone recordings, supporting scalable preliminary assessment and telehealth applications in veterinary cardiology.

Motivation & Objective

• Address the lack of real-world, smartphone-recorded canine heart sound data for AI-based analysis.
• Develop and validate a robust multi-stage SoNUS 3.2.x pipeline with quality gating for variable recording conditions.
• Provide heart rate estimation benchmarks and assess feasibility for at-home telecardiology in veterinary practice.

Proposed method

• Collect a real-world canine heart sound dataset (100+ dogs, 5 continents) using smartphone microphones under routine conditions.
• Annotate 38 recordings by board-certified veterinary cardiologists for S1/S2 timing, murmurs, ectopic beats, and arrhythmias.
• Develop SoNUS 3.2.x with a multi-stage fallback architecture and a quality scoring module.
• Create a fast 30–40s variant and a primary 60s variant to evaluate short- and long-duration performance.
• Train/finetune a murmur detection model (ResNet) on PhysioNet CinC datasets and adapt to canine data.

Experimental results

Research questions

• RQ1 Can smartphone-recorded canine heart sounds provide reliable heart rate estimates under real-world recording conditions?
• RQ2 Does a multi-stage fallback and a quality gating mechanism improve robustness to recording variability?
• RQ3 What are the performance characteristics of a fast 30–40s model versus a primary 60s model for HR estimation in dogs?
• RQ4 Is murmur detection tractable in opportunistic home-recorded canine heart sounds after domain adaptation?

Key findings

• Fast model (30–40s) achieves mean HR error 11.14%, mean HR accuracy 88.86%, and median HR accuracy 92.98%.
• Primary model (60s) achieves mean HR error 8.37%, mean HR accuracy 91.63%, and median HR accuracy 94.95%.
• Quality scoring (QS) filtering (threshold ≥70) improves percentile accuracy and reduces unreliable outputs, with 86% of short recordings and 84.21% of full-length recordings deemed suitable for results presentation.
• Quality scoring filtered results show higher 80th/90th percentile HR accuracies when QS ≥70 (e.g., 80th percentile 84.82-88.71; 90th percentile 76.49-86.29).
• SoNUS 3.2.x’s multi-stage fallback enhances robustness to variable recording quality, enabling recovery of valid results from lower-quality inputs.
• Murmur detection achieved state-of-the-art benchmark performance on PhysioNet CinC datasets but real-world murmur evaluation was limited by signal quality and ambient noise.