Bioelectronic Medicine (Nov 2023)

Rib detection using pitch-catch ultrasound and classification algorithms for a novel ultrasound therapy device

  • Claire R. W. Kaiser,
  • Adam B. Tuma,
  • Maryam Zebarjadi,
  • Daniel P. Zachs,
  • Anna J. Organ,
  • Hubert H. Lim,
  • Morgan N. Collins

DOI
https://doi.org/10.1186/s42234-023-00127-0
Journal volume & issue
Vol. 9, no. 1
pp. 1 – 13

Abstract

Read online

Abstract Background Noninvasive ultrasound (US) has been used therapeutically for decades, with applications in tissue ablation, lithotripsy, and physical therapy. There is increasing evidence that low intensity US stimulation of organs can alter physiological and clinical outcomes for treatment of health disorders including rheumatoid arthritis and diabetes. One major translational challenge is designing portable and reliable US devices that can be used by patients in their homes, with automated features to detect rib location and aid in efficient transmission of energy to organs of interest. This feasibility study aimed to assess efficacy in rib bone detection without conventional imaging, using a single channel US pitch-catch technique integrated into an US therapy device to detect pulsed US reflections from ribs. Methods In 20 healthy volunteers, the location of the ribs and spleen were identified using a diagnostic US imaging system. Reflected ultrasound signals were recorded at five positions over the spleen and adjacent ribs using the therapy device. Signals were classified as between ribs (intercostal), partially over a rib, or fully over a rib using four models: threshold-based time domain classification, threshold-based frequency domain classification, logistic regression, and support vector machine (SVM). Results SVM performed best overall on the All Participants cohort with accuracy up to 96.25%. All models’ accuracies were improved by separating participants into two cohorts based on Body Mass Index (BMI) and re-fitting each model. After separation into Low BMI and High BMI cohorts, a simple time-thresholding approach achieved accuracies up to 100% and 93.75%, respectively. Conclusion These results demonstrate that US reflection signal classification can accurately provide low complexity, real-time automated onboard rib detection and user feedback to advance at-home therapeutic US delivery.

Keywords