Using machine learning to automate ultrasound-based classification of butt-fused joints in medium-density polyethylene gas pipes.

Polyethylene (PE) pipes are widely used in gas distribution. Their joints are prone to various flaws and are the most problematic part of the pipeline, so the infrastructure industry requires an effective inspection technique. Butt-fusion (BF) is the most common method of joining PE pipes. In this research, we investigated the applicability of machine learning (ML) to automate the ultrasonic inspection of PE pipe BF joints. Flawless and defective joints were fabricated. A-scan signals were collected from each group of samples using a customized chord transducer, with the aim of developing and assessing the viability of ML approaches to the problem of joint classification. We compared several ML approaches to the problem and found that convolutional neural networks were most performant, classifying signals with an F1 score of 0.874 in a four-class problem (identifying defect presence and type) and of 0.912 in binary classification (defect presence/absence only). Our results show that an ultrasonic chord-type transducer approach can effectively resolve flawless samples versus those with coarse contaminants or cold fusions and that an ML approach can be used to effectively assess these ultrasonic signals. Our findings can be used to develop a portable, efficient, user-friendly, and inexpensive device for in-field joint inspections.

High-resolution quantitative acoustic microscopy of cutaneous carcinoma and melanoma: Comparison with histology.

BACKGROUND: The increased incidence rate of skin cancers during the last decades is alarming. One of the significant difficulties in the histopathology of skin cancers is appearance variability due to the heterogeneity of diseases or tissue preparation and staining process. This study aims to investigate whether the high-resolution acoustic microscopy has the potential for identifying and quantitatively classifying skin cancers. MATERIAL/METHODS: Unstained standard formalin-fixed skin tissue samples were used for ultrasonic examination. The high-frequency acoustic microscope equipped with the 320 MHz transducer was utilized to visualize skin structure. Fourier transform was performed to calculate the sound speed and attenuation in the tissue. RESULTS: The acoustic images demonstrate good concordance with the traditional histology images. All histological features in the tumour were easily identifiable on acoustic images. Each skin cancer type has its combination of ultrasonic properties significantly different from the healthy skin. CONCLUSIONS: High-resolution acoustic imaging strengthened with quantitative analysis shows a potential to work as an auxiliary imaging modality assisting pathologists to lean to the particular decision in doubtful cases. The method can also assist surgeon to ensure the complete resection of a tumour.

An Ultrasonic Adaptive Beamforming Method and Its Application for Trans-Skull Imaging of Certain Types of Head Injuries; Part II: Reception Mode and Adaptive Imaging.

OBJECTIVE: Background theory and a new algorithm for single-point adaptive focusing in transmission mode through ultrasonic barriers via one-dimensional phased arrays were reported in part I. In this paper the algorithm is further extended and implemented into a full adaptive beamforming process, including complete transmission and reception modes. METHODS: Corrected time delay patterns, adapted to the local acoustical and geometrical properties of the barrier, are calculated and applied in both modes. Further, an adaptive imaging process is also developed that implements the proposed beamforming process for two-dimensional imaging through randomly shaped multilayered phase-aberrating structures. The method is optimized for the case of human skull as the ultrasound barrier and its application for transcranial imaging is discussed. RESULTS: Laboratory results of adaptive imaging through realistic skull-mimicking phantoms are presented. The algorithms are implemented on a 64-channel ultrasound open-source phased array platform controlling a standard 128-element biomedical phased array. Irregularly shaped reflectors with characteristic dimensions of the order of ∼0.5 mm to ∼4.5 mm were used as targets behind the skull phantoms in our experiments. Minimum and maximum distortional target displacements of 2.2 mm and 25.3 mm (in 12 cm depth) were observed in sonograms when uncompensated time delays were used. By contrast, the positioning errors ranged from 0.0 to 0.9 mm when our algorithm was employed. CONCLUSION AND SIGNIFICANCE: The adaptive imaging results demonstrate strong potential of the proposed technique for diagnostic imaging of acoustically reflective head injuries directly through intact human skull.

Estimating the parameters of audible clinical percussion signals by fitting exponentially damped harmonics.

Used for centuries in the clinical practice, audible percussion is a method of eliciting sounds by tapping various areas of the human body either by finger tips or by a percussion hammer. Despite its advantages, pulmonary diagnostics by percussion is still highly subjective, depends on the physician's skills, and requires quiet surroundings. Automation of this well-established technique could help amplify its existing merits while removing the above drawbacks. In this work, clinical percussion signals from normal volunteers are decomposed into a sum of exponentially damped sinusoids (EDS) whose parameters are determined using the Matrix Pencil Method. Some EDS represent transient oscillation modes of the thorax/abdomen excited by the percussion event, while others are associated with the noise. It is demonstrated that relatively few EDS are usually enough to accurately reconstruct the original signal. It is shown that combining the frequency and damping parameters of these most significant EDS allows for efficient classification of percussion signals into the two main types historically known as "resonant" and "tympanic." This classification ability can provide a basis for the automated objective diagnostics of various pulmonary pathologies including pneumothorax. The algorithm can be implemented on an embedded platform for the battlefield and other emergency applications.

Accurate 3-D Profile Extraction of Skull Bone Using an Ultrasound Matrix Array.

The present study investigates the feasibility, accuracy, and precision of 3-D profile extraction of the human skull bone using a custom-designed ultrasound matrix transducer in Pulse-Echo. Due to the attenuative scattering properties of the skull, the backscattered echoes from the inner surface of the skull are severely degraded, attenuated, and at some points overlapped. Furthermore, the speed of sound (SOS) in the skull varies significantly in different zones and also from case to case; if considered constant, it introduces significant error to the profile measurement. A new method for simultaneous estimation of the skull profiles and the sound speed value is presented. The proposed method is a two-folded procedure: first, the arrival times of the backscattered echoes from the skull bone are estimated using multi-lag phase delay (MLPD) and modified space alternating generalized expectation maximization (SAGE) algorithms. Next, these arrival times are fed into an adaptive sound speed estimation algorithm to compute the optimal SOS value and subsequently, the skull bone thickness. For quantitative evaluation, the estimated bone phantom thicknesses were compared with the mechanical measurements. The accuracies of the bone thickness measurements using MLPD and modified SAGE algorithms combined with the adaptive SOS estimation were 7.93% and 4.21%, respectively. These values were 14.44% and 10.75% for the autocorrelation and cross-correlation methods. Additionally, the Bland-Altman plots showed the modified SAGE outperformed the other methods with -0.35 and 0.44 mm limits of agreement. No systematic error that could be related to the skull bone thickness was observed for this method.

A one-dimensional numerical model of acoustic wave propagation in a multilayered structure of a resistance spot weld.

A one-dimensional model of acoustic wave propagation in a multilayered structure of a spot weld is developed. The inhomogeneity of the material properties due to the thermal inhomogeneity is included in the equation of motion. The model enables us to deal with arbitrary spatial distributions of Lamé constants and density. The model allows analysis of travel time, multiple reflections, and interference in a given geometry. Use of this model could provide information to help predict behavior of the waves in the transmission (reflection) mode at different plate thicknesses and welding settings.

Ultrasonic imaging of static objects through an aberrating layer using harmonic phase conjugation approach.

The main goal of this study is to develop a new image reconstruction approach for the ultrasonic detection of small objects (comparable to or smaller than the ultrasonic wavelength) behind an aberrating layer. Instead of conventional pulse-echo experimental setup we used through transmission, as the backscattered field after going twice through the layer becomes much weaker than the through-transmitted field. The proposed solution is based on the Harmonic Phase Conjugation (HPC) technique. The developed numerical model allows to calculate the amplitude and phase distributions of the through-transmitted acoustic field interacting with the objects and received by a linear transducer array either directly or after passing through an additional aberrating layer. Then, the digitized acoustic field received by the array is processed, phase-conjugated, and finally, numerically propagated back through the medium in order to reconstruct the image of the target objects. The reconstruction quality of the algorithm was systematically tested on a numerical model, which included a barrier, a medium behind it, and a group of three scatterers, by varying scatterer distances from the source transducer, their mutual arrangement, and the angle of the incident field. Subsequently, a set of laboratory experiments was conducted (at transmit frequency of 2 MHz) to verify the accuracy of the developed simulation. The results demonstrate feasibility of imaging multiple scattering objects through a barrier using the HPC method with better than 1mm accuracy. The results of these tests are presented, and the feasibility of implementing this approach for various biomedical and NDT imaging applications is discussed.

Assessment of gingival thickness using an ultrasonic dental system prototype: A comparison to traditional methods.

Knowledge of periodontal anatomy is essential when performing surgical and non-surgical procedures in the field of oral healthcare. Gingival thickness (GT) is often assessed for this purpose. A dental system prototype was recently developed for quantitative, non-invasive GT assessment by high-frequency (HF) ultrasound. Laboratory trials were conducted to validate system performance against a traditional method of assessment. A system with a 50 MHz broadband, spherically-focused transducer was used. The transducer was housed in a small, hand-held probe equipped with a continuous water supply. A-scans were obtained and thickness at each location was determined. For comparison, the traditional method of transgingival probing through tissue with an endodontic k-file needle was also implemented. Preliminary experiments were performed on phantoms simulating the anatomical and acoustic properties of human periodontal tissues. A porcine cadaver was obtained for further laboratory trials. The speed of sound through porcine gingiva was determined to be 1564 ± 21 m/s. Finally, a multiple-point experiment involved GT assessment in an array of locations on the buccal gingival surface in the fourth quadrant. Ultrasonic measurements were found to yield similar GT values to those obtained from invasive methods. Results obtained in this experiment validate the applicability of ultrasound as a diagnostic tool for assessing periodontal anatomy.

An Ultrasonic-Adaptive Beamforming Method and Its Application for Trans-skull Imaging of Certain Types of Head Injuries; Part I: Transmission Mode.

A new adaptive beamforming algorithm for imaging via small-aperture 1-D ultrasonic-phased arrays through composite layered structures is reported. Such structures cause acoustic phase aberration and wave refraction at undulating interfaces and can lead to significant distortion of an ultrasonic field pattern produced by conventional beamforming techniques. This distortion takes the form of defocusing the ultrasonic field transmitted through the barrier and causes loss of resolution and overall degradation of image quality. To compensate for the phase aberration and the refractional effects, we developed and examined an adaptive beamforming algorithm for small-aperture linear-phased arrays. After accurately assessing the barrier's local geometry and sound speed, the method calculates a new timing scheme to refocus the distorted beam at its original location. As a tentative application, implementation of this method for trans-skull imaging of certain types of head injuries through human skull is discussed. Simulation and laboratory results of applying the method on skull-mimicking phantoms are presented. Correction of up to 2.5 cm focal point displacement at up to 10 cm depth under our skull phantom is demonstrated. Quantitative assessment of the method in a variety of temporal focusing scenarios is also reported. Overall temporal deviation on the order of a few nanoseconds was observed between the simulated and experimental results. The single-point adaptive focusing results demonstrate strong potential of our approach for diagnostic imaging through intact human skull. The algorithms were implemented on an ultrasound advanced open-platform controlling 64 active elements on a 128-element phased array.

New data on histology and physico-mechanical properties of human tooth tissue obtained with acoustic microscopy.

Quantitative evaluation of human tooth structural elements, revealed in acoustic images, has been carried out. It has been shown that tissue elements with different acoustic impedances differed in acoustic images by intensity of grey color, and also feature with different longitudinal sound velocities (C(L)). In the layer of mantle dentin, C(L) is 7% to 8% lower than in bulk dentin, and in the layer of dentin around the pulp chamber, C(L) is 15% lower. In carious enamel and dentin, C(L) decreases up to 7% to 17%. In pathologic teeth, dentin areas with higher density can be revealed; they feature higher C(L); in transparent dentin C(L) can be 15% to 20% higher than in bulk dentin. Results of the present study show that acoustic images reflect internal biomechanical properties of tooth tissue microstructure that can be evaluated quantitatively by means of longitudinal sound velocity determination.

An ultrasonic array technique for material characterization of plate samples.

An ultrasonic system with a linear array for characterization of a layered specimen placed in immersion liquid parallel to the aperture of the array is considered. To estimate the longitudinal and transverse wave velocities as well as the thickness and density of the specimen, it is proposed to decompose the spatio-temporal data recorded by the array in a spectrum of plane pulse waves. Based on fitting the developed wave model of the system to the experimental data, it is shown that the relative delays and amplitudes of the spectral responses can be used for the estimation of the velocities and thickness of the layer and its density. The distortions of the plane wave spectrum caused by the spatial discretization of the array data are considered. It is proposed to suppress these distortions using individual interpolating processing of the received pulses separated in the spatio-temporal domain. The developed technique is experimentally verified on a fused quartz plate evaluated with a 17-MHz linear array. The relative reproducibility of the estimation is found to be 0.11% in the longitudinal wave velocity and thickness of the plate, and 0.5% and 5% in the transverse wave velocity and the density, respectively.

Advantages and limitations of using matrix pencil method for the modal analysis of medical percussion signals.

Although clinical percussion remains one of the most widespread traditional noninvasive methods for diagnosing pulmonary disease, the available analysis of physical characteristics of the percussion sound using modern signal processing techniques is still quite limited. The majority of existing literature on the subject reports either time-domain or spectral analysis methods. However, Fourier analysis, which represents the signal as a sum of infinite periodic harmonics, is not naturally suited for decomposition of short and aperiodic percussion signals. Broadening of the spectral peaks due to damping leads to their overlapping and masking of the lower amplitude peaks, which could be important for the fine-level signal classification. In this study, an attempt is made to automatically decompose percussion signals into a sum of exponentially damped harmonics, which in this case form a more natural basis than Fourier harmonics and thus allow for a more robust representation of the signal in the parametric space. The damped harmonic decomposition of percussion signals recorded on healthy volunteers in clinical setting is performed using the matrix pencil method, which proves to be quite robust in the presence of noise and well suited for the task.

Enamel thickness measurement with a high frequency ultrasonic transducer-based hand-held probe for potential application in the dental veneer placing procedure.

This study presents a novel approach to measure the enamel thickness potentially applicable to the veneer placing procedure. All experiments have been carried out on the extracted human teeth, using a high frequency ultrasonic transducer (50 MHz, Sonix, Springfield, VA, USA). The enamel thickness measurement results obtained with high positional accuracy by a scanning acoustic microscope (Tessonics AM1103, Windsor, ON, Canada) were compared with the measurements conducted in a hand-held mode by using the same transducer placed in a custom fixture. Finally, to validate the ultrasonic measuring results, the samples were cut down the long axis to expose the cross-section. The enamel thickness was measured in several points along the selected part of the exposed cross-section by using an optical microscope (Stemi SV 11, Carl Zeiss AG, Jena, Germany). The values of the enamel thickness received by using the hand-held probe vs. the acoustic microscope were in close proximity (~10% difference) and were also satisfactory close to the enamel thickness results obtained from the direct cross-sectional measurements (~12% difference). The authors suggested a measuring procedure that allows avoiding errors related to the ultrasonic beam localization on the tooth surface. The high feasibility of the ultrasonic pulse-echo measurements in a hand-held mode was demonstrated.

Acoustic imaging of thick biological tissue.

Up to now, biomedical imaging with ultrasound for observing a cellular tissue structure has been limited to very thinly sliced tissue at very high ultrasonic frequencies, i.e., 1 GHz. In this paper, we present the results of a systematic study to use a 150 to 200 MHz frequency range for thickly sliced biological tissue. A mechanical scanning reflection acoustic microscope (SAM) was used for obtaining horizontal cross-sectional images (C-scans) showing cellular structures. In the study, sectioned specimens of human breast cancer and tissues from the small intestine were prepared and examined. Some accessories for biomedical application were integrated into our SAM (Sonix HS-1000 and Olympus UH-3), which operated in pulse-wave and tone-burst wave modes, respectively. We found that the frequency 100 to 200 MHz provides optimal balance between resolution and penetration depth for examining the thickly sliced specimens. The images obtained with the lens focused at different depths revealed cellular structures whose morphology was very similar to that seen in the thinly sectioned specimens with optical and scanning acoustic microscopy. The SAM operation in the pulse-echo mode permits the imaging of tissue structure at the surface, and it also opens up the potential for attenuation imaging representing reflection from the substrate behind the thick specimen. We present such images of breast cancer proving the method's applicability to overall tumor detection. SAM with a high-frequency tone-burst ultrasonic wave reveals details of tissue structure, and both methods may serve as additional diagnostic tools in a hospital environment.

A study of the adhesion between dental cement and dentin using a nondestructive acoustic microscopy approach.

OBJECTIVES: The goal of the present study was to investigate the potential for acoustic microscopy techniques to characterize the cement-dentin interface in restored teeth. METHODS: Special flat-parallel specimens and whole extracted teeth with restorations were scanned using a high-frequency (50 MHz) focused ultrasonic transducer. Visual acoustic images (B- and C-scans) of the cement-dentin interface were obtained nondestructively, analyzed and compared with optical images taken after the samples were cut along the scanning axis. The shear bonding strength of a subsection of specimens was tested in a Lloyd material testing machine. RESULTS: An essential distinction between the acoustical properties associated with good and failed bonding has been shown. In the case of failed adhesion, the ultrasound signal reflection from the cement-dentin interface is up to four to seven times higher in magnitude than in the case of good bonding. The comparison of the ultrasound imaging data with the data obtained using an optical microscope revealed a strong correspondence with the acoustical and optical results with respect to the presence, position and dimensions of the defects. The specimens showing higher ultrasound reflection from cement/dentin interface have also shown lower shear bonding strength. SIGNIFICANCE: The results demonstrate that acoustic scanning with a high-frequency focused ultrasonic probe is a valuable method for nondestructive morpho-mechanical analysis of cement/dentin interface for either experimental models or whole restored teeth. An appropriately expanded approach can be widely used for the pre-clinical evaluation of dental materials. Further, this method may prove beneficial in the design of new diagnostic ultrasound devices and techniques for use within clinical dentistry.

Pulse-echo NDT of adhesively bonded joints in automotive assemblies.

A new method for the detection of void-disbonds at the interfaces of adhesively bonded joins is considered. Based on a simple plane wave model, the output waveform is presented as a sum of two responses associated with the reflection of the ultrasonic wave at the first metal-adhesive interface and the second metal-adhesive interface, respectively. The strong response produced by the wave reverberating in the first metal sheet is eliminated through comparison between the pulse-echo signal measured at the area under the test and reference waveform recorded for the bare first metal sheet outside of the joint. The developed decomposition algorithm has been applied to the study of steel and aluminum samples having various adhesive layer thicknesses in a range of 0.1-1mm.

Application of a nonlinear boundary condition model to adhesion interphase damage and failure.

In an earlier paper [J. Sadler, B. O'Neill, and R. Maev, J. Acoust. Soc. Am. 118, 51-59 (2005)], a set of generalized boundary conditions were proposed, based on a thin layer (thickness << wavelength) model of the acoustic interface. In this paper, the model is extended to cover the more pathological nonlinearity of the adhesion interphase-that is, the critically important thin layer where bonds are formed between adhesive and substrate. First, the boundary conditions are shown to be sufficiently general to cope with all manner of interphase nonlinearity, including unilateral cases such as clapping or slipping. To maintain this generality, an analytic time domain solution is proposed based on expansion in terms of the layer thickness rather than the conventional expansion in terms of harmonics. Finally, the boundary conditions are applied to an interphase failure model based upon basic continuum damage mechanics principles. It is proposed that such a model, which can predict the evolution of the interphase damage under stressful conditions, may allow a proper prediction of the ultimate adhesion strength based on nonlinear parameters measured nondestructively with ultrasound.

Ultrasonic wave propagation across a thin nonlinear anisotropic layer between two half-spaces.

Boundary conditions and perturbation theory are combined to create a set of equations which, when solved, yield the reflected and transmitted wave forms in the case of a thin layer of material that is perfectly bonded between two isotropic half-spaces. The set of perturbed boundary conditions is created by first using the fully bonded boundary conditions at each of the two interfaces between the thin layer and the half-spaces. Then, by restricting the layer's thickness to be much smaller than an acoustic wavelength, perturbation theory can be used on these two sets of boundary equations, producing a set of equations which effectively treat the thin layer as a single interface via a perturbation term. With this set of equations, the full range of incident and polar angles can be considered, with results general enough to use with a layer that is anisotropic, nonlinear, or both anisotropic and nonlinear. Finally the validity of these equations is discussed, comparing the computer simulation results of this theory to results from standard methods, and looking at cases where the results (or various properties of the results) are known or can be predicted.