pulse of US generated from absorbers everywhere in the exposed 3D tissue volume propagate towards a specially designed acoustic lens of focal length f . The lens enables the simultaneous focusing of all the waves on the other side of the lens. If the center of the tissue is kept at a distance 2f from the lens, then a 32 element linear array of US transducers can be placed at 2f distance on the other side to detect the focused PA signals at 32 different pixel locations in the image plane for each laser firing. These US time signals, referred to as A-line signals, were amplified and then digitized at 30 MHz on 32 independent channels simultaneously. A-line signals were envelope detected in order to keep only the slowly varying nonzero signal values. The linear array was scanned in the image plane with repeated laser firing to collect PA signals over the entire image plane. The spatial resolution achieved by this system was around 1.3 mm . PA signals from different depth planes along the lens axis arrive at the image plane at different arrival times due to the finite propagation speed of the US in water. By taking time slices on all the A-line signals, one can generate 2D C-scan PA images that correspond to different depth planes in the tissue as in figure (8.5). We were able to pick a time gate of 200 sample width, indicated by t 1 and t 1 in figure (8.5), that included all PA signals coming from the
Neuschmelting et al. evaluated both 2D and 3D handheld MSOT (multispectral optoacoustic tomogr- aphy) devices developed by iThera Medical by imaging murine brain melanomas at 4 MHz and 700 – 900 nm . The authors implanted B16F10 melanoma cells into the right frontal lobe, where tumors 3.5 – 7.0 mm in diameter grew within 13 days; the size of these tumors were independently measured with MRI. The 2D system was deemed more accurate (bias = 0.24 mm) than the 3D system (bias = 2.35 mm) but was limited to axial tumor depths of 4 mm. The 3D probe was limited by tumor size (< 3.5 mm) in all dimensions, but the acquisition time was shorter. The performance disparity between the 2D and 3D systems was attributed to a non-ideal distribution of diffused light in the transducer array of the 3D system. These handheld systems also showed superi- or signal to noise ratios and limits of detection and for metastatic melanoma cells (5 cells/uL at 8 mm tissue depths in mice) to PET/CT using fluorine 18 fluoro- deoxyglucose . However, this sensitivity would be difficult to achieve if applied to human lymph nodes located centimeters below the skin. Additionally, to our knowledge, these systems have not yet been used to monitor resection during or after an operation.
As the spectrum of Photoacoustic (PA) radio frequency (RF) signals contain significant information about the structure of PA absorbers, parameters extracted from PA spectrum can be potentially used to differentiate between different tissue pathologies. Quantitative reconstruction of chromophore concentration or optical properties of soft tissue from the multispectral PA image pixels is very difficult as the pixel values depend on several factors like the optical absorption and scattering inside the tissue, frequency dependent acoustic attenuation etc. Although there are several methods proposed by different research groups for quantitative recovery of absorption property and chromophore concentration, some of the methods can be applied to only specific cases i.e. their application is limited by the assumption on which they have been built, while the more general methods have been tested only in simulation [5.7]. An alternative approach for tissue characterization using PA data is the analysis of the PA frequency spectrum as it carries significant information about the shape and size of the PA absorbers [5.2]. 5.2.1 Literature Review
Liver cancer or hepatic cancer is a cancer that originates in the liver. Liver is the largest internal organ. Leading cause of the liver cancer is cirrhosis due to either hepatitis B virus (HBV) or hepatitis C virus (HCV) . Liver cancer is common in countries in sub-Saharan Africa and Southeast Asia. In many of these countries, it is the most common type of cancer. More than 700,000 people are diagnosed with this type of cancer each year throughout the world. Liver cancer is also the leading cause of cancer deaths worldwide, accounting for more than 600,000 deaths each year . Cancer diagnosis begins with a through physical examination. However, liver cancer is often hard to be diagnosed this way because signs and symptoms often do not appear until the tumor becomes quite large . When a tumor is suspected, imaging tests such as X- ray, Computed Tomography (CT), Magnetic
To these ends, we adapted the spatiotemporal mean- shift (STM-S) approach to our context, which demon- strated good performance for magnetic resonance image clustering by taking into account both spatial and evo- lutive features . We have already shown some pre- liminary result on a single experimental dataset , and further to this, we present here an automatic pre- processing pipeline and a deeper study of the parame- ter choices. Moreover, the quantitative validation is fully explained, and the performance of STM-S is evaluated on several photoacoustic datasets that were acquired from two different home-made phantoms. The first phantom contained two pieces of the same medium that have to be discriminated, one of which was fresh tissue and the other of which was stale tissue. The second phantom contained two different dilutions of the same biological medium, i.e., blood, which corresponded to two differ- ent concentrations of oxygen inside these inclusions that could be clustered together or separately, depending on the application. Indeed, it might be important to discrim- inate different concentrations of a single medium, e.g., to determine the concentration of a contrast agent in the body, and to discriminate a single medium from all other media without considering their dilutions, e.g., to deter- mine the level of vascularization for calculation of the oxygenation rate .
in phantoms and in vitro could be highly suitable for in vivo deep tissue PA imaging. In addition, its absorbance spectrum differs significantly from those of the major background tissue chromophores such as oxy- and deoxyhemoglobin (Figure S1), facilitating accurate multispectral unmixing of individual PA signals. To test the feasibility of applying folate-CP dots for noninvasive PA molecular imaging of folate receptors, they were injected intravenously into mice bearing FR + ve MCF-7 breast cancer xenografts. CP dots with no folate functional- ization were used as the control. PA signals were collected at five different excitation wavelengths – 700, 750, 810, 850, and 900 nm – based on the absorption maxima and minima of the probe, and oxy- and deoxyhemoglobin at 1, 3, 5, 24, 48, 72, and 168 hours postadministration. To compensate for the attenuation of excitation light in deep tissues, the images were corrected using a wavelength-dependent exponential decay model, which assumes a uniform distribution of absorption and scattering properties, as well as constant tissue oxygenation. Subsequently, the images were spectrally decomposed to reveal the chromophore-specific contrast using known extinc- tion spectra. Figure 4B–C are the MSOT maximum intensity projection images of the transverse and coronal abdomen sec- tions of the tumor-bearing mice at 3 hours post folate-CP dots injection and Figure 4D–E are those injected with CP dots. PA signals at all other time points are provided in Figure S3. Figure 4F is obtained from ROI analyses on single-slice images corresponding to the tumor site at various time points after probe injection. A strong PA signal could be detected as early as 1 hour in mice injected with folate-CP dots, which reached a maximum at 3 hours (Supplementary video) after which it gradually dropped, whereas the PA signal from the tumor in
As light travels deeper into the tissue, tight optical focusing is no longer achievable due to multiple scattering events of photons . Correspondingly, for an acoustic-resolution PA system, weakly “focused” light is usually used to excite a significantly larger volume of the sample than that in an optical-resolution PA system (Figure 1b, 1d). However, as acoustic waves are much less scattered than photons in biological tissue, acoustic focusing is superior to optical focusing beyond a specific imaging depth and will yield a lateral resolution that is determined by the ultrasound focusing . The lateral resolution of acoustic-resolution PA systems is typically within hundreds of microns to tens of microns (Table 1), which is scalable with the centre frequency of the ultrasound transducer. Benefitting from the large imaging depth, the acoustic-resolution PA system is more commonly used in whole-body imaging applications (Table 1). Wang et al. applied AR-PAM system in conjunction with a novel targeted indocyanine green-doped nanoprobe to image breast cancer in mice (Figure 5) . For AR-PACT systems, three types of detection geometries have been reported: planar , circular  and spherical [29, 30]. Wang et al. presented a three-dimensional volumetric PA imaging system built on a two-dimensional planar matrix array ultrasound probe and demonstrated the clinical potential of the system to identify sentinel lymph nodes for cancer staging purposes (Figure 6) . To perform whole-body imaging, Xia et al. reported a novel small animal whole-body PA imaging system with a confocal design of free-space ring-shaped light illumination and 512-element full-ring ultrasonic array signal detection (Figure 7) . Kruger et al.
Multispectral Infrared imaging is a simple non-invasive technique that has been widely applied in Cultural Heritage analysis and study [1-5]. In its simplest realization, four images of the subject under study are acquired in the spectral bands of Blue, Green, Red and Infrared; in most cases, the Infrared image carries the most relevant infor- mation, because infrared radiation penetrates under the surface, thus allowing for the visualization of otherwise in- visible details such as underdrawings and ‘pentimenti’ in canvas and panel paintings [6,7]. Infrared imaging is also important for other applications, because of the possible enhancement of features deriving from the different infra- red reflectivities of the subject’ s constituent materials. The improvement of readability of degraded manuscripts in the Infrared image was demonstrated, for example, in the re- covery of the burnt Erculaneum scrolls . However, in some cases, useful information can be derived by the ana- lysis of the whole multispectral series; in a recent paper  we presented a new approach based on blind signal pro- cessing [10-12] for extracting hidden information from a painting. The guiding idea is that the appearance of the
A multispectralimaging system captures data within the specified region of interest across the electromagnetic spectrum. The multispectral image produced is usually created with multiple bands of wavelengths collected using narrow-band LED illumination or colored filters placed between the object and the sensor (Berns, R. S., 2019). Each channel yields spectral information that is uniquely present only in the specified wavelength region. For example, a certain pigment in an artwork cannot be detected in the visible spectrum but can be detected in the infrared region. As such, increasing the number of channels yields more spectral information (Wang, Y., & Berns, R. S. , 2017; Dickinson, C. , 2001). Bear in mind that creating such systems increases cost and complexity. Berns (2005) discussed a trade-off between spectral accuracy, colorimetric accuracy, cost, and hardware and software complexity. Most multispectralimaging systems use monochrome cameras. An ideal multispectralimaging system would have good colorimetric accuracy, good spectral accuracy, and low image noise (Berns, Roy S., 2005).
Chapter 5 Table 5.1.Comparison of PSNR for a skin cancer image after simulating different noises and de-noising by filters in different densities……………………………………………… 59 Table 5.2. The most effective Filters on different noises with densities between 10% - 80%................................................................................................................................... 60 Table 5.3 .Result table of Modified Hausdorff Distance……………………………..…… 61
Briefly, TIL generation was performed similarly to that outlined by Surgery Branch, NCI, NIH, tiltum protocol 9- 6-05 (provided by DR. John Wunderlich). Typically, at least 6 wells of tumor fragments and 6 wells enzymatically digested tumor were plated for culture. For tumor frag- ment cultures, typically 4 to 10 fragments were plated into each well of a 24 well plate containing 2 mL human AB CM supplemented with 1000 cU/mL. Enzyme digested tumor suspensions were adjusted to 5.0 × 10 5 /mL in human AB CM supplemented with 1000 cU/ml IL-2 and 2mls were plated per well of a 24 well plate. The 24 well plates were placed in a humidified 37 °C incubator with 5 % CO2 and cultured until lympho- cyte growth was evident. Each well of the plate was inspected on alternate days using a low-power inverted microscope to monitor the extrusion and proliferation of lymphocytes. Whether or not lympho- cyte growth was visible, half of the medium was replaced in all wells no later than 1 week after culture initiation. Typically, about 1 to 2 weeks after culture initiation, a dense lymphocytic carpet would cover a portion of the plate surrounding each fragment. When any well became almost confluent, all the growing lymphocytes were mixed vigorously, split into two daughter wells and filled to 2 mL per well with CM plus 1000 cU/mL IL-2. Subsequently, the cultures were split to maintain a cell density of 0.5 to 1.0 × 10 6 cells/mL, or half of the media was replaced at least twice weekly or (as needed). The age of TIL cultures used in these studies varied from 12 to 67 days. Each cul- ture originating from each of the initial 6 wells plated for each condition were considered to be an independent TIL culture or “cloid” and maintained in a separate plate with separate pipettes used to maintain integrity of each cloid during expansion.
calibration location, clearly indicates the directionality of each transducer. For example, transducer 12 and 13 (columns 12 and 13) stare at the object space from an orientation off the acoustic axis. These results correlate well with the actual physical orientation of the transducers indicating the calibration map is accurately reporting signal sensitivity. Of interest is the extreme sensitivity reduction outside the axial line-of-sight for all transducers. This implied that signal generated and transmitted from a photoacoustic source off the acoustic axis (approximately ± 15 mm) could be easily concealed in the background noise. This suggests that the large 25 mm diameter transducers used to populate the transducer array confined the effective object space to a spherical volume 30x30x30 mm 3 by the overlap of the calibration sensitivity maps. Figure 2.4(b) illustrates the behavior of the signal FWHM at each calibration location. The difference in FWHM is generally small from position to position. However, the outer edges of the object space tend to broaden the acoustic signal as it sweeps over the face of the transducer as opposed to directly impinging the transducer when the signal propagates on the acoustic axis. While visibly difficult to discern, this trend is generally evident in the FWHM patterns in that they tend to show larger signals at the periphery of a given z-slice. As well, the complexity of the FWHM patterns indicates any modeling to predict the FWHM of a PA signal would be very difficult. Shown in Fig. 2.4(c) is the time-of-flight map. Perhaps the most important information garnered from the time-of-flight response is the presence of dark spots at a variety of locations in the object space. Typically these positions were found in regions which corresponded to low signal amplitude (as seen in Fig. 2.4(a)). Likely, signals were masked in the noise and went undetected at those particular calibration positions because of the reduced amplitudes off the acoustic axis of a transducer.
gether. Doyle et al. [86, 87, 88] developed a cascaded ensemble learning system dividing the multiclass problem into several binary problems, going from the broadest to the most specific. Tissues are first sorted out between cancerous and non-cancerous. Then, the cancerous tissues are subdivided according to another binary classifier – e.g. Gleason grades 3 and 4 versus Gleason grade 5. This same approach is used until all the classes are processed. Through this process, the most different classes are better divided and it results in an increased accuracy. This method has proved to outperform the traditional one-versus-all scheme used for multiclass problems and the one-shot classifi- cation. The overall multiclass accuracy was 89 %. Nguyen et al.  used two SVM classifiers trained on different feature sets, meaning one of them is trained on texture features and the other is trained on morphological features. The probabilities that each classifier classifies their associated feature set as cancer or normal tissue are multiplied. Those products are then compared and the sample is classified as cancerous if the product of the probabilities that it is cancerous is greater than the product of the probabilities that it is normal tissue. Greenbalt et al.  presented a two-stage ensemble learning system that first assigns an initial grade using quaternion wavelets and LPB associated with a neural network multiclass classification. In a second phase, the classification result is refined using a SVM classifier if some classes have close probabilities. An accuracy of 98.9 % was reported over all the classes considered. Such a system can be generalised to more than two stages using a tree-like structure.
Abstract—This paper presents a simple feed-forward back- propagation Neural Network (NN) model to detect and locate early breast cancer/tumor eﬃciently through the investigation of Electro- magnetic (EM) waves. A spherical tumor of radius 0.25 cm was cre- ated and placed at arbitrary locations in a breast model using an EM simulator. Directional antennas were used to transmit and receive Ultra-Wide Band (UWB) signals in 4 to 8 GHz frequency range. Small training and validation sets were constructed to train and test the NN. The received signals were fed into the trained NN model to ﬁnd the presence and location of tumor. Very optimistic results (about 100% and 94.4% presence and location detection rate of tumor respectively) have been observed for early received signal components with the NN model. Hence, the proposed model is very potential for early tumor detection to save human lives in the future.
In this paper, we propose a novel image reconstruction algorithm that incorporates the finite bandwidth charac- teristics of ultrasound transducers. An optimal filter is designed to deconvolve the transducer’s impulse re- sponse of the finite bandwidth at each imaging point. Through the numerical simulation, the proposed algo- rithm is compared to those reconstruction algorithms assuming finite-bandwidth ultrasound transducers.
In , we used UWB signals in time domain. The signals were sampled to obtain the feature vector needed to train the NN model. PulsOn device does not show the time relation between the transmitted and the received signals, it only shows the wave form. So, it is impossible to sample the received signals at the same instance of time. The DCT is a sum of infinitely many cosines basis functions in different frequencies with varying real number amplitudes . By this summation, DCT can approximate any signal. Since most of the signals information is captures in few low frequency components, we have decided to use DCT. Applying DCT to the same data obtained in  gives tumor existence, size and location detection rate of 100%, 86.9% and 80.4% respectively. A simple back-propagation feed- forward NN model was used. The input feature vector size is 200. It corresponds to the DCT components between 50 and 250. It has one hidden layer of 4 nodes and an output layer with one node for tumor location detection. The same NN architecture is used for tumor size detection. Tumor existence detection is noticed by the negative output in both NN architectures. MatLab 7 was used for calculating DCT and constructing the NN model
two days. The DMEM medium was then replaced with 2 ml of the same medium containing each AuNR type at a concentra- tion of 1 10 11 NP ml 1 . Aer 4 h incubation, the AuNR- medium was removed and the cell monolayer on the cover- slip was twice-rinsed with DPBS (14190-094, Life Technolo- gies, UK), xed in 4% paraformaldehyde/DPBS for 10 min at room temperature and rinsed with DPBS twice. The xed coverslips were mounted and sealed onto glass slides. Bright and dark-eld microscopy imaging was performed with an inverted microscope (Nikon Eclipse Ti-E, Nikon UK Ltd, UK) and an oil coupled 100 objective (CFI Plan Fluor, Nikon UK Ltd, UK). Images were recorded with a 5 Megapixel colour camera (DS-Fi1, Nikon UK Ltd, UK) and saved using the NIS- Elements D soware (Nikon UK Ltd, UK). Open-source so- ware package ImageJ 69 was used to crop and enhance the
For the first time, we report photoacoustic (PA) signal detection in a cell placed within the Mi- chelson interferometer cavity in an attempt to relate photoacoustic effect to the Michelson fringe shift as a result of changes in the cell. Both detection schemes were investigated using IR absorp- tion and their sensitivities compared. Signals related to Michelson interferometer fringe and PA effect have shown good correlations with each other using different samples including some es- sential oils and their corresponding plant part from which the essential oil is usually obtained. Results were encouraging and will open the door widely to use the combined Michelson interfe- rometer-photoacoustic spectroscopy (PAS) in trace gas detection for different applications.
In this paper we have presented a compressive sampling based imaging and unmixing scheme for multispectral data processing,which does not require any a priori knowledge of the endmember distribution. The method has applications in ﬂuorescent imaging of small animals. Through the use of compressive sampling, the requirements on hardware design are less stringent than for traditional approaches. Currently, the work of this paper applies to when the signal of interest is comparable to the autoﬂuorescence and the demand for resolution is not high, due to the denoising method of removing pixels under chosen threshold for background noise induced by compressive sampling. The numerical results presented in this paper clearly demonstrate the potential of this method to be able to extract essential spectral informationin a precise manner. Extending this technique to situations with low signal to noise ratio is theoretically achievable since NESTA and MVSA have the capability to operate at high accuracy [12, 16], but the denoising procedure would require more sophistication. Compressive sampling unmixing, as a complement to standard unmixing techniques, has the potential to be applied in the real large-scale multispectralimaging applications in the future.
In vivo and ex vivo PA experiments We conducted in vivo experiments on rats to confirm that NiPNP with strong absorbance at 1064 nm could be used as the PA agents in deep tissues (Figures 4–6). We imaged clinically important SLN (n = 3), GI tract (n = 3) and cystography (n = 3) . We adopted the rat model because the skin of a rat is more structurally similar to human tissue than other rodents. Figure 4 shows the PA/US images of SLNs, Figure 5 shows those of GI tracts, and Figure 6 shows those of bladders. Panels A–C in Figure 4–6 represents the PA maximum amplitude projection (MAP) images acquired pre-NiPNP injection, post-NiPNP injection without chicken tissues, and post-NiPNP injection with chicken tissues, respectively. Panels D–F of Figure 4–6 are obtained by applying a depth encoded image processing method to the panels A–C, respectively. Panels G–I in Figure 4–6 represent the 2D depth-resolved PA/US images obtained along the white dashed lines in the panels A–C, respectively. Panel J in Figure 4–6 shows the quantification results of PA amplitude enhancement at the NiPNP injection sites (pre-injection, post-injection without chicken tissues, and post-injection with chicken tissues). The imaging region is indicated by the black dashed boxes in the animal photographs as shown in Figure 4K–6K. The PA amplitude enhancement was calculated as the percentage increase in the PA amplitudes pre- and post-NiPNP injection (i.e., (PA after –PA before )/ PA before ×