SOMA : Certifying Depth Reliability Limit of a Passive-Acoustic Tissue Stiffness Classifier

Background and Objective Passive acoustic sensing offers a potential approach to non-invasive tissue-stiffness assessment. However, signal attenuation with propagation depth, combined with sensor noise and acquisition variability, can undermine classification reliability even when a model continues to produce predictions. This study develops a statistical framework for identifying the depth at which a single-channel passive-acoustic tissue-stiffness classifier becomes indistinguishable from chance. Using liver-stiffness classification as its application context, the work investigates the influence of model training and selected physical and acquisition parameters on this reliability boundary. Methodology In the absence of a suitable public dataset pairing depth-resolved acoustic signals with stiffness labels, the investigation uses simulated shear-wave signals generated through a Kelvin–Voigt viscoelastic attenuation model. Tissue stiffness is varied across three classes spanning 3–15 kPa, with sensor noise and selected acquisition effects incorporated into the simulation. The principal experiment evaluates 24 independently trained one-dimensional convolutional neural networks across 36 propagation depths. At each depth, network performance is compared against an empirical null distribution generated from 719 retrained shuffled-label controls, with statistical significance assessed using Benjamini–Hochberg correction. Additional experiments examine training duration, classifier behaviour, timing jitter, and class-specific reliability. Principal Findings The 200-epoch CNN experiment produces a mean certified operational floor of 13.9 cm (SD = 1.1 cm), with an empirical 10th-percentile value of 12.3 cm. Under otherwise paired conditions, reducing training duration to 60 epochs lowers the mean floor to 9.3 cm (SD = 3.7 cm). Further experiments demonstrate sensitivity to acquisition timing and reveal that class-specific degradation can precede the overall reliability boundary. Scope and Limitations The reported results are simulation-derived and do not establish real-world penetration limits or clinical performance. The model assumes homogeneous tissue, and certain hardware and acquisition parameters remain approximations. A separate phantom-signal experiment verifies pipeline processing but does not validate stiffness classification or the predicted depth boundary. Experimental validation using characterised hardware, depth-resolved recordings, and layered tissue models remains necessary.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-15
DOI
https://doi.org/10.5281/zenodo.24248064
Primary Topic
Ultrasound Imaging and Elastography
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

SOMA : Certifying Depth Reliability Limit of a Passive-Acoustic Tissue Stiffness Classifier

Riddhit Majumdar
Zenodo (CERN European Organization for Nuclear Research)
Ultrasound Imaging and Elastography
preprint

SOMA : Certifying Depth Reliability Limit of a Passive-Acoustic Tissue Stiffness Classifier

Riddhit Majumdar
preprint en

Abstract

Background and Objective Passive acoustic sensing offers a potential approach to non-invasive tissue-stiffness assessment. However, signal attenuation with propagation depth, combined with sensor noise and acquisition variability, can undermine classification reliability even when a model continues to produce predictions. This study develops a statistical framework for identifying the depth at which a single-channel passive-acoustic tissue-stiffness classifier becomes indistinguishable from chance. Using liver-stiffness classification as its application context, the work investigates the influence of model training and selected physical and acquisition parameters on this reliability boundary. Methodology In the absence of a suitable public dataset pairing depth-resolved acoustic signals with stiffness labels, the investigation uses simulated shear-wave signals generated through a Kelvin–Voigt viscoelastic attenuation model. Tissue stiffness is varied across three classes spanning 3–15 kPa, with sensor noise and selected acquisition effects incorporated into the simulation. The principal experiment evaluates 24 independently trained one-dimensional convolutional neural networks across 36 propagation depths. At each depth, network performance is compared against an empirical null distribution generated from 719 retrained shuffled-label controls, with statistical significance assessed using Benjamini–Hochberg correction. Additional experiments examine training duration, classifier behaviour, timing jitter, and class-specific reliability. Principal Findings The 200-epoch CNN experiment produces a mean certified operational floor of 13.9 cm (SD = 1.1 cm), with an empirical 10th-percentile value of 12.3 cm. Under otherwise paired conditions, reducing training duration to 60 epochs lowers the mean floor to 9.3 cm (SD = 3.7 cm). Further experiments demonstrate sensitivity to acquisition timing and reveal that class-specific degradation can precede the overall reliability boundary. Scope and Limitations The reported results are simulation-derived and do not establish real-world penetration limits or clinical performance. The model assumes homogeneous tissue, and certain hardware and acquisition parameters remain approximations. A separate phantom-signal experiment verifies pipeline processing but does not validate stiffness classification or the predicted depth boundary. Experimental validation using characterised hardware, depth-resolved recordings, and layered tissue models remains necessary.

Zenodo (CERN European Organization for Nuclear Research)
Ultrasound Imaging and Elastography
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon