Photoacoustic imaging (PAI) has the potential to detect cancer in the early stage. PAI is safe due to its non-ionizing radiation properties, which greatly enhance its clinical feasibility in the near future, which provides significant benefits over other imaging techniques like X-ray computed tomography (CT). In this paper, the fully automated 3D deep learning cancer detector is taken to detect and localize the presence of cancer in freshly excised ex-vivo human thyroid and prostate tissue specimens using a three-dimensional (3D) multispectral photoacoustic (MPA) dataset automatically. The model detected and localized the cancer region in a given test MPA image with promising results.
Multispectral photoacoustic (MPA) specimen imaging modality is proven successful in differentiating photoacoustic (PA) signal characteristics from a cancer and normal region. The oxy and de-oxy hemoglobin content in a human tissue captured in the MPA data are the key features for cancer detection. In this study, we propose to use deep 3D convolution neural network trained on the thyroid MPA dataset and tested on the prostate MPA dataset to evaluate this potential. The proposed algorithm first extracts the spatial, spectral, and temporal features from the thyroid MPA image data using 3D convolutional layers and detects cancer tissue using the logistic function, the last layer of the network. The model achieved an AUC (area under the curve) of the ROC (receiver operating characteristic) curve of 0.72 on the prostate MPA dataset.
Pathology diagnosis is usually done by a human pathologist observing tissue stained glass slide under a microscope. In the case of multi-specimen study to locate cancer region, such as in thyroidectomy, significant labor-intensive processing is required at high cost. Multispectral photoacoustic (MPA) specimen imaging, has proven successful in differentiating photoacoustic (PA) signal characteristics between a histopathology defined cancer region and normal tissue. A more pragmatic research question to ask is, can MPA imaging data predict, whether a sectioned tissue slice has cancer region(s)? We propose to use inception-resnet-v2 convolutional neural networks (CNNs) on the thyroid MPA data to evaluate this potential by transfer learning. The proposed algorithm first extracts features from the thyroid MPA image data using CNN and then detects cancer using the softmax function, the last layer of the network. The AUCs (area under curve) of the receiver operating characteristic (ROC) curve of cancer, benign nodule and normal are 0.73, 0.81, and 0.88 respectively with a limited number of the MPA dataset.
This paper evaluates the detection performance of the three subpixel target detection algorithms based on the spectral signature of a target. Three subpixel target detection algorithms, Adaptive Coherence Estimator (ACE), Spectral Matched Filter (SMF), and Constrained Energy Minimization (CEM) are evaluated and compared using the Principal Component Analysis (PCA) spaced RIT Avon12 hyperspectral dataset. The performance of the three detectors is evaluated by generating the Receiver Operating Characteristic (ROC) curve. The ROC curves are generated by uploading the detection statistics image produced by the three detectors to the Data and Algorithm Standard Evaluation ( DASE) Website of IEEE Geoscience and Remote Sensing Society(GRSS) . Finally, we note the Area Under Curve (AUC) as the proposed utility metric value to evaluate the performance of the three detectors. The AUCs of the ROC curve produced by the ACE, CEM, and SMF are 94.0 %, 93.9 %, and 87.2 % respectively.
Intracranial hemorrhage is a critical conditional with the high mortality rate that is typically diagnosed based on head computer tomography (CT) images. Deep learning algorithms, in particular, convolution neural networks (CNN), are becoming the methodology of choice in medical image analysis for a variety of applications such as computer-aided diagnosis, and segmentation. In this study, we propose a fully automated deep learning framework which learns to detect brain hemorrhage based on cross sectional CT images. The dataset for this work consists of 40,367 3D head CT studies (over 1.5 million 2D images) acquired retrospectively over a decade from multiple radiology facilities at Geisinger Health System. The proposed algorithm first extracts features using 3D CNN and then detects brain hemorrhage using the logistic function as the last layer of the network. Finally, we created an ensemble of three different 3D CNN architectures to improve the classification accuracy. The area under the curve (AUC) of the receiver operator characteristic (ROC) curve of the ensemble of three architectures was 0.87. Their results are very promising considering the fact that the head CT studies were not controlled for slice thickness, scanner type, study protocol or any other settings. Moreover, the proposed algorithm reliably detected various types of hemorrhage within the skull. This work is one of the first applications of 3D CNN trained on a large dataset of cross sectional medical images for detection of a critical radiological condition
KEYWORDS: Continuous wave operation, Pulsed laser operation, Signal generators, Photoacoustic spectroscopy, MATLAB, Signal to noise ratio, Ultrasonography, Real time imaging, Optical simulations, Semiconductor lasers
Photoacoustic (PA) imaging is a hybrid imaging modality that integrates the strength of optical and ultrasound imaging. Nanosecond (ns) pulsed lasers used in current PA imaging systems are expensive, bulky and they often waste energy. We propose and evaluate, through simulations, the use of a continuous wave (CW) laser whose amplitude is linear frequency modulated (chirp) for PA imaging. The chirp signal provides signal-to-side-lobe ratio (SSR) improvement potential and full control over PA signal frequencies excited in the sample. The PA signal spectrum is a function of absorber size and the time frequencies present in the chirp. A mismatch between the input chirp spectrum and the output PA signal spectrum can affect the compressed pulse that is recovered from cross-correlating the two. We have quantitatively characterized this effect. The k-wave Matlab tool box was used to simulate PA signals in three dimensions for absorbers ranging in size from 0.1 mm to 0.6 mm, in response to laser excitation amplitude that is linearly swept from 0.5 MHz to 4 MHz. This sweep frequency range was chosen based on the spectrum analysis of a PA signal generated from ex-vivo human prostate tissue samples. In comparison, the energy wastage by a ns laser pulse was also estimated. For the chirp methodology, the compressed pulse peak amplitude, pulse width and side lobe structure parameters were extracted for different size absorbers. While the SSR increased 6 fold with absorber size, the pulse width decreased by 25%.
KEYWORDS: Acoustics, Prostate cancer, Photoacoustic imaging, Photoacoustic spectroscopy, Cancer, Signal detection, Signal to noise ratio, Tumors, Prostate, Near infrared
There is an urgent need for sensitive and specific tools to accurately image early stage, organ-confined human prostate cancers to facilitate active surveillance and reduce unnecessary treatment. Recently, we developed an acoustic lens that enhances the sensitivity of photoacoustic imaging. Here, we report the use of this device in conjunction with two molecular imaging agents that specifically target the prostate-specific membrane antigen (PSMA) expressed on the tumor cell surface of most prostate cancers. We demonstrate successful imaging of phantoms containing cancer cells labeled with either of two different PSMA-targeting agents, the ribonucleic acid aptamer A10-3.2 and a urea-based peptidomimetic inhibitor, each linked to the near-infrared dye IRDye800CW. By specifically targeting cells with these agents linked to a dye chosen for optimal signal, we are able to discriminate prostate cancer cells that express PSMA.
We have designed and implemented a novel acoustic lens based focusing technology into a prototype photoacoustic imaging camera. All photoacoustically generated waves from laser exposed absorbers within a small volume get focused simultaneously by the lens onto an image plane. We use a multi-element ultrasound transducer array to capture the focused photoacoustic signals. Acoustic lens eliminates the need for expensive data acquisition hardware systems, is faster compared to electronic focusing and enables real-time image reconstruction. Using this photoacoustic imaging camera, we have imaged more than 150 several centimeter size ex-vivo human prostate, kidney and thyroid specimens with a millimeter resolution for cancer detection. In this paper, we share our lens design strategy and how we evaluate the resulting quality metrics (on and off axis point spread function, depth of field and modulation transfer function) through simulation. An advanced toolbox in MATLAB was adapted and used for simulating a two-dimensional gridded model that incorporates realistic photoacoustic signal generation and acoustic wave propagation through the lens with medium properties defined on each grid point. Two dimensional point spread functions have been generated and compared with experiments to demonstrate the utility of our design strategy. Finally we present results from work in progress on the use of two lens system aimed at further improving some of the quality metrics of our system.
Frequency domain analysis of the photoacoustic (PA) radio frequency signals can potentially be used as a tool for characterizing microstructure of absorbers in tissue. This study investigates the feasibility of analyzing the spectrum of multiwavelength PA signals generated by excised human prostate tissue samples to differentiate between malignant and normal prostate regions. Photoacoustic imaging at five different wavelengths, corresponding to peak absorption coefficients of deoxyhemoglobin, whole blood, oxyhemoglobin, water and lipid in the near infrared (NIR) (700 nm – 1000 nm) region, was performed on freshly excised prostate specimens taken from patients undergoing prostatectomy for biopsy confirmed prostate cancer. The PA images were co-registered with the histopathology images of the prostate specimens to determine the region of interest (ROI) corresponding to malignant and normal tissue. The calibrated power spectrum of each PA signal from a selected ROI was fit to a linear model to extract the corresponding slope, midband fit and intercept parameters. The mean value of each parameter corresponding to malignant and adjacent normal prostate ROI was calculated for each of the five wavelengths. The results obtained for 9 different human prostate specimens, show that the mean values of midband fit and intercept are significantly different between malignant and normal regions. In addition, the average midband fit and intercept values show a decreasing trend with increasing wavelength. These preliminary results suggest that frequency analysis of multispectral PA signals can be used to differentiate malignant region from the adjacent normal region in human prostate tissue.
Acoustic lens based focusing technology where the image reconstruction is achieved through the focusing of an acoustic lens, can potentially replace time consuming and expensive electronic focusing technology for producing high resolution real time ultrasound (US) images. A novel acoustic lens focusing based pulse echo US imaging system is explored here. In the system, a Polyvinylidene fluoride (PVDF) film transducer generates plane wave which is backscattered by the object and focused by a spherical acoustic lens on to a linear array of transducers. To improve the anticipated low signal to noise ratio (SNR) of the received US signal due to the low electromechanical coupling coefficient of the PVDF film, here we explored the possibility of implementing pulse compression technique using linear frequency modulated (FM) signals or chirp signals. Comparisons among the different SNR values obtained with short pulse and after pulse compression with chirp signal show a clear improvement of the SNR for the compressed pulse. The preliminary results show that the SNR achieved for the compressed pulse depends on time bandwidth product of the input chirp and the spectrum of the US transducers. The axial resolution obtained with compressed pulse improved with increasing sweep bandwidth of input chirp signals, whereas the lateral resolution remained almost constant. This work demonstrates the feasibility of using a PVDF film transducer as an US transmitter in an acoustic lens focusing based imaging system and implementing pulse compression technique into the same setup to improve SNR of the received US signal.
Prostate cancer is the second leading cause of death in American men after lung cancer. The current screening
procedures include Digital Rectal Exam (DRE) and Prostate Specific Antigen (PSA) test, along with Transrectal
Ultrasound (TRUS). All suffer from low sensitivity and specificity in detecting prostate cancer in early stages. There is a
desperate need for a new imaging modality. We are developing a prototype transrectal photoacoustic imaging probe to
detect prostate malignancies in vivo that promises high sensitivity and specificity. To generate photoacoustic (PA)
signals, the probe utilizes a high energy 1064 nm laser that delivers light pulses onto the prostate at 10Hz with 10ns
duration through a fiber optic cable. The designed system will generate focused C-scan planar images using acoustic lens
technology. A 5 MHz custom fabricated ultrasound sensor array located in the image plane acquires the focused PA
signals, eliminating the need for any synthetic aperture focusing. The lens and sensor array design was optimized
towards this objective. For fast acquisition times, a custom built 16 channel simultaneous backend electronics PCB has
been developed. It consists of a low-noise variable gain amplifier and a 16 channel ADC. Due to the unavailability of 2d
ultrasound arrays, in the current implementation several B-scan (depth-resolved) data is first acquired by scanning a 1d
array, which is then processed to reconstruct either 3d volumetric images or several C-scan planar images. Experimental
results on excised tissue using a in-vitro prototype of this technology are presented to demonstrate the system capability
in terms of resolution and sensitivity.
New analysis tools to address the problem of early detection of the eye blinding disease glaucoma are presented. The thickness maps of the Retinal Nerve Fiber Layer (RNFL) corresponding to 184 eyes (92 Normal and 92 Glaucoma Patients) were obtained from a Scanning Laser Polarimeter (Gdx-VCC). The two dimensional data was used to draw features as opposed to the circular band one-dimensional data in previous approaches. Fourier analysis was performed on the 90° projection of the thickness map data to emphasize the shape contained in the RNFL. Different parameters from the Fourier Coefficients were drawn and tested for their ability to detect glaucoma. Significant differences were found in the shape measures of the projections and the ROC curve analysis was done to measure the separability of the sample set with those features. Another approach was to analyze the shape of the entire 2 dimensional thickness map through a 2D Fourier Transform. A circular ring band (10 pixel wide) data at a radius of 20 pixels was analyzed for this 2D FT. Principal Component Analysis was performed on this data for dimension reduction of feature space. Finally Fisher's linear discriminant function (LDF) was used as a classifier. The evaluation of different parameters obtained through the Fourier analysis of the thickness map image of RNFL was found to be a useful tool as an analysis strategy for glaucoma detection.
Ultrasound speckle carries information about the interrogated scattering microstructure. The complex signal is represented as a superposition of signals due to all scatterers within a resolution cell volume, VE. A crossbeam geometry with separate transmit and receive transducers is well suited for such studies. The crossbeam volume, VE is defined in terms of the overlapping diffraction beam patterns. Given the focused piston transducer's radius and focal distance, a Lommel diffraction formulation suitable for monochromatic excitation is used to calculate VE as a function of frequency and angle. This formulation amounts to a Fresnel approximation to the diffraction problem and is not limited to the focal zone or the far field. Such diffraction corrections as VE are needed to remove the system effects when trying to characterize material using moment analysis. Theoretically, VE is numerically integrated within the overlapping region of the product of the transmit-receive transfer functions. Experimentally, VE was calculated from the field pattern of a medium-focused transducer excited by a monochromatic signal detected by a 0.5mm diameter PVDF membrane hydrophone. We present theoretical and experimental evaluations of VE for the crossbeam geometry at frequencies within the transducers' bandwidth, and its application to tissue microstructure characterization.
In medical ultrasonic imaging the signal reflected from the tissue often has a random character to it. It is believed that the random nature of the tissue scattering microstructure is responsible for the stochastic nature of the echo signal. Chen, et. al. Have proposed a signal processing scheme that is based on the statistical moments calculated on the Fourier transform of the time gated echo signal. The theory requires the knowledge of a frequency- dependent effective cell volume term. This paper describes the use of a closed form expression (Lommel diffraction formulation) for this purpose. Our simulation results suggest that reliable estimation of the cell volume is possible only when the time duration of the excitation pulse is small compared to the time gate length.
KEYWORDS: Medical imaging, Transducers, Fermium, Imaging systems, Tissues, Point spread functions, Frequency modulation, Scattering, Ultrasonography, Signal processing
Ultrasonic echoes, backscattered from an inhomogeneous medium have the character of a random signal, which is mainly responsible for the observed speckle in medical images. Such a medium can be modeled as a uniform matrix with scattering bodies distributed randomly. When the number of density of scatterers is high, the individual scatterers are not resolved by the imaging process, and a speckle pattern is produced as a result of interference of waves from many scatterers within the resolution cell volume. This cell volume depends on the beam profile and the pulse width of the interrogating pulse. The authors have used a 3D simulation phantom that takes into account the 3D distribution of scatterers and the 3D nature of the resolution cell volume. Several simulations were performed to study the effect of scatterer number density (SND) and resolution cell volume on the backscattered signal. Assuming the process is linear and the stochastic signal is ergodic and stationary, Kurtosis (K), which involves 2nd and 4th moments, was estimated in each case. It was found that Kurtosis varies linearly with another parameter Fs that depends on the resolution cell volume. The results are analyzed in the light of theoretical predictions. Reasonable estimates of SND can be derived from the slope of Kurtosis vs. parameter Fs graph.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.