Open Access
Add to BibSonomy
Abstract
In our previous studies, we have developed a prototype interstitial needle sensing probe that can acquire broadband A-line photoacoustic (PA) signals encoding both tissue microarchitecture and histochemical information comparable to that accessible by histology. Paving the road toward clinical translation of this technology, we replaced the piezoelectric hydrophone in the needle PA probe with a fiber optic hydrophone that enabled both broader bandwidth and sufficient signal-to-noise ratio (SNR) for PA signal detection. Such an all-optical design also facilitated disposability and significantly reduced the footprint of the needle PA sensing probe. Experiments were performed on well-controlled phantoms and human prostate tissues. The microarchitectures in each sample were quantitatively evaluated by both the nonlinear spectral slope of the PA signal power spectrum and the generalized gamma (GG) parameter a by implementing envelope statistics to the PA signal. In the studies on phantoms containing optically absorbing microspheres with various sizes and concentrations, the nonlinear spectral slope showed a strong correlation of r=-0.80 with the microsphere dimensions, and a relatively weak correlation of r=-0.54 with the microsphere concentrations, while the GG parameter a showed a strong correlation with the microsphere dimensions (r=0.72) and a moderate correlation with the microsphere concentrations (r=0.63). In the studies on human prostate tissues containing progressive cancer stages, both the nonlinear spectral slope and the GG parameter a demonstrated a statistically significant difference between benign and nonaggressive cancer tissues (p<0.01), and between nonaggressive and aggressive cancer tissues (p<0.01). In addition, a multivariate analysis combining the two quantitative measurements demonstrated the boundaries among the different progressive stages of prostate cancer.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Tissue microarchitecture plays a critical role in the pathological diagnosis of diseases [1]. For example, the microarchitecture of prostate glands is evaluated by pathologist and assigned with Gleason scores that are the current clinical standard for evaluating the prognosis for patients with prostate cancer. The current standard procedure for assessing tissue microarchitecture is biopsy followed by histopathology which involves invasive tissue extraction [2,3], places the patient at risk of biopsy-related complications [4], and contributes to expense and time for histological processing. Therefore, non-invasive, or minimally invasive methods for instantaneous in vivo assessment of tissue microarchitecture are highly desired.
Recently developed photoacoustic (PA) spectral analysis (PASA) method, by analyzing the power spectrum of PA signal in the frequency domain, offers a new method for quantitative assessment of the microarchitecture in optically absorbing biological tissues [5,6]. By further taking advantage of the unique optical absorption spectra of individual histochemical components in tissues, such as hemoglobin, lipid, and collagen, PASA of multi-wavelength PA measurements of a tissue can assess tissue microarchitectures formed by each histochemical component. This noninvasive method facilitates the characterization of tissue microarchitecture and histochemical properties simultaneously [7–9]. Multi-wavelength PASA has been validated in previous studies and adapted to several applications, including quantifying progressive architectural changes in prostate cancer [9,10], identifying ovarian cancer [11], characterizing non-alcoholic fatty liver disease [8,12], differentiating intraocular tumor types [13], determining red blood cell aggregation [14], and characterizing disease within lymph nodes [15].
As demonstrated in our previous studies, the capability of PASA in characterizing micron scale structures, e.g. morphological changes of red blood cells [16] and the clustering pattern of cancer cells [10], is dominantly relevant to the bandwidth of the detected PA signal. Highly sensitive detection of high frequency PA signal from deep tissue in vivo, however, can be challenging as the attenuation coefficient of PA signals is generally proportional to the square of the signal frequency [17].
In trade of high resolution and quantitative microarchitecture information comparable to histology, we sacrificed the 2D imaging capability using an ultrasound transducer array outside the tissue-volume-of-interest and developed a prototype interstitial needle PA sensing probe [9,18]. This probe consists of a fiber optic diffuser with approximately 10 mm illumination length and a needle hydrophone with broad bandwidth up to 30 MHz. When inserted in a minimally invasive manner, the needle PA probe can capture the high frequency A-line PA signal within illuminated tissue volume. With sufficient temporal length and bandwidth, the microarchitecture information can be extracted with statistics-based spectral analysis.
The design of the needle PA sensing probe is aligned with the standard needle biopsy procedures but aimed at minimizing tissue extractions. The fiber optic diffuser has low manufacture cost and is disposable after single use, although the emission energy is not perfectly uniform along the fiber optic diffuser [9,18]. The hydrophone has limited receiving aperture and its sensitivity is affected by nonuniform frequency domain attenuation. These factors, in combination, form a sensitivity distribution with low frequency spatial variation. Such low frequency spatial variation contributes to a constant low frequency component in any measurement that is not considered in PASA. The sampling volume of the needle PA sensing probe is defined by the threshold contour within which sufficient SNR can be achieved. The limitation of sampling volume can be compensated by the insertions of the needle PA probe at multiple locations within the prostate, which is not achievable with traditional 1D or 2D transducer arrays for external imaging. Sampling at multiple locations is commonly performed in needle biopsy procedures. For example, prostate lesion is clinically significant only if its diameter is larger than 10 mm [19]. Therefore, with multiple insertions separated by 10 mm, a clinically significant cancerous lesion in the prostate can be identified. We validated the performance of the prototype interstitial needle PA probe in rodent prostates in vivo [20], and canine [18] and human prostates ex vivo [9].
Our previously developed needle PA sensing probe used a piezo-electric hydrophone for PA signal detection. Although this hydrophone possesses excellent detection sensitivity, it has a relatively large diameter of approximately 3 mm. In addition, the high manufacturing cost of the piezo-electric hydrophone makes the interstitial needle PA probe expensive and non-disposable. In this study, for the first time to the best of our knowledge, we produced an interstitial needle PA sensing probe with an all-optical design. The piezo-electric hydrophone for PA signal detection was replaced by a commercially available fiber optic hydrophone. Compared to the piezo-electric hydrophone, the fiber optic hydrophone carries the advantage of a small footprint for miniaturization of the needle PA sensing probe. For the particular case of prostate cancer diagnosis, the broader signal detection bandwidth of the fiber optic hydrophone not only covers the previously observed clustered cancer cells, but also the epithelium with smaller dimensions [20]. After compensated by the frequency response of the hydrophone, the signal power spectra may demonstrate peaks contributed by each of the two components, leading to a nonmonotonic and nonlinear shape. Fitting such power spectra with linear model [5,6] may not be sufficient for comprehensively characterizing the complicated tissue microarchitecture. Hence, we introduced by a nonlinear spectral analysis method. In addition, we followed previous studies [21] and implemented an envelope statistics method for time domain analysis.
The performance of the all-optical needle PA sensing probe powered by the updated analysis methods for tissue characterization was validated first using phantoms containing optically absorbing microarchitectures with various dimensions and densities. Then, the diagnostic capability of the all-optical needle PA sensing probe was examined via the experiments on human prostate tissues with different stages of progressive prostate cancer.
2. Methods
2.1 Experiment setup
As illustrated in Fig. 1(a), the illumination sources of the experiment system include an optical parametric oscillator (OPO) pumped by the second harmonic output of a pulsed Nd:YAG laser (690-950 nm and 1200-2400 nm, 5-7 ns pulse width, 10 Hz repetition rate, OPO Mobile, OPOTEK, Carlsbad, CA) and the fourth harmonic output of an Nd:YAG laser (266 nm, 5-8 ns pulse width, 10Hz repetition rate, Surelite, Continuum, San Jose, CA). The all-optical needle PA sensing probe, as shown in Fig. 1(b), consists of a fiber optic diffuser and a fiber optic hydrophone. The fiber optic diffuser, when coupled with the laser output, delivers a cylindrical illumination pattern, as shown in Fig. 1(c). The fiber optic diffuser was fabricated following a fully automated protocol [22]. In brief, the plastic jacket of a 1000 µm-diameter optic fiber was removed within 12 mm from its emission end. The emission end was positioned in a 50 mL centrifuge tube without touching the tube wall. Hydrofluoric acid ≤ 48% (HF, Sigma-Aldrich, Cleveland, OH) was infused into the tube at a flow rate of 13 µL/min. The HF acid surface gradually rose over the fiber emission end and etched a 10 mm-long conical surface. At the coupling end of the fiber optic diffuser, the optical energy was 27 mJ at 1220 nm and 1370 nm, and 10 mJ at 266 nm. The optical energy density on the surface of the fiber optic diffuser was approximately 51 mJ/cm2 at 1220 nm and 1370 nm. This energy level is much lower than the safety limit of 100 mJ/cm2 at these wavelengths established by American National Standard Institute (ANSI). At 266 nm, the optical density on the surface of the fiber optic diffuser was 12 mJ/cm2 that was higher than the ANSI safety limit of 4 mJ/cm2 at 266 nm. This, however, is not a concern for ex vivo tissue samples.
Fig. 1. All-optical needle PA sensing probe setup. (a) System schematics. FOD: Fiber optic diffuser, FOH: Fiber optic hydrophone. FOHS: Fiber optic hydrophone system. (b) The needle PA sensing probe. (c) A photo showing the emission pattern of the fiber optic diffuser. (d) Intensity distribution along fiber diffusor needle in Fig. 1(c). 0 mm and 10mm represent the positions a and b labeled in Fig. 1(c), respectively. (e) Frequency response of the fiber optic hydrophone.
By measuring the emission power with an optical power meter through a 1 mm slit that was moved along the longitudinal dimension of the fiber optic diffuser, the emission optical intensity distribution along the fiber optic diffuser was quantified, as shown in Fig. 1(d). The PA signals generated by the fiber optic diffuser were received by a fiber optic hydrophone (OD=150 µm, Precision Acoustics, UK) [23] positioned parallel to the fiber optic diffuser, as illustrated in Fig. 1(b). The acquired signals were displayed on an oscilloscope and stored by a computer after amplification by 20 dB (5072 Pulse/Receiver, Olympus, Center Valley, PA). All signals were calibrated using the frequency response of the fiber optic hydrophone shown in Fig. 1(e).
2.2 Spatial sensitivity distribution of the all-optical needle PA sensing probe
We quantified the spatially distributed sensitivity offered by the all-optical needle PA sensing probe. A black microsphere with a diameter of 1 mm (BKPMS-1.2, Cospheric LLC., Santa Barbara, California) was fixed at the tip of a positioning fiber. Driven by a 2D translational stage, the microsphere was positioned at different locations with respect to the needle probe. Both the needle probe and the point object were submerged in 1% intralipid solution, which has been frequently used for mimicking the optical properties of biological tissue [24,25]. The microsphere was moved 15 mm along the longitudinal direction (z direction) and 4 mm along the azimuthal direction (x direction), both at a step size of 0.25 mm, as the scanning area shown by the dashed rectangle in Fig. 2(a). At each position in the scanning area, PA signals from the microsphere were measured, and the SNR were quantified by the peak-to-peak amplitude of the PA signal from the microsphere over the averaged amplitude of the background noise. Then the spatially distributed sensitivity was presented as the SNR as a function of the microsphere locations. The same procedure was repeated at three different laser wavelengths, including 1220 nm, 1370 nm, and 266 nm, which correlate to the strong absorption of lipid, collagen, and nucleus acid, respectively.
Fig. 2. Measurement geometries. (a) Measuring spatial sensitivity distribution of the all-optical needle probe using 1 mm diameter microsphere as a point object. Driven by 2D translation stages, the point object fixed at the tip of a positioning fiber was placed at different locations in the dashed rectangle area. D: diameter. (b) A photo taken during a phantom study. (c) A photo taken during the study on a human prostate tissue. The dashed lines in (b) and (c) are the emission segments of the fiber optic diffusers.
2.3 Study on phantoms
We further investigated the capability of the all-optical needle PA sensing probe in distinguishing the dimensions and concentrations of optically absorbing microspheres by repeating our previous study using the piezo-electric hydrophone [5]. As shown in Fig. 2(b), cylindrical phantoms were fabricated with polyethylene microspheres (UVPMS-BV, Fisher scientific LLC., Santa Barbara, California) uniformly and randomly distributed in 10% porcine gel (G2500, Sigma-Aldrich Co. LLC., Burlington, MA). Illumination at 532 nm was used to target the dark purple color of the microspheres. We fabricated 9 phantoms, including 5 containing 100 µm diameter microspheres at different concentrations of 2, 6, 10, 12, and 20 microspheres per mm3, and 4 containing 40, 50, 100, and 200 µm diameter microspheres at the concentration of 2 microspheres per mm3. For each phantom, 6 measurements were taken at separated locations in the phantom, resulting in a total of 54 independent measurements.
2.4 Study on human prostate tissues
Via the study on human prostate tissues, we examined the performance of the all-optical needle PA sensing probe in differentiating progressive cancer stages. Tissue samples with known pathological results were obtained from the Specialized Programs of Research tissue core at the University of Michigan. The tissue procurement protocol was approved by the institutional review board of human subject research at the University of Michigan. The sample pool included 7 benign human prostate tissue samples, 4 nonaggressive cancer human prostate tissue samples [Gleason score (GS) = 6], and 3 aggressive cancer human prostate tissue samples (GS=7). The representative histology photographs of the samples are shown in Fig. 3. As shown in Fig. 2(c), each tissue block had the dimensions of approximately 15 mm×10 mm×3 mm. To prevent possible contamination of the tissue samples that were to be returned to the tissue core, we inserted the fiber optic diffuser into the sample, covered the tissue sample with optimal cutting temperature compound, and sealed the sample with saran wrap. The PA measurements in each sample were acquired at 2 to 3 different locations with separations of 5 mm on average. Due to the limited spatial detection range shown later in the results section, we considered the measurements from these 2-3 different locations to be independent. The total number of measurement locations was 30, including 14 in benign tissues, 8 in nonaggressive tissues, and 8 in aggressive tissues. At each location, measurements at the three laser wavelengths, including 1220 nm, 1370 nm, and 266 nm, were taken.
Fig. 3. Representative histology photographs of the measured human prostate tissue samples, including a benign sample, a nonaggressive cancer tissue with GS=6, and an aggressive cancer tissue with GS=7. 100× objective magnification.
2.5 Methods for quantitative analysis of tissue microarchitecture
The fiber optic hydrophone enables a broader detection bandwidth [23] compared to the piezoelectric needle hydrophone in our previous studies [5], which highly benefits the capability of the needle PA probe in characterizing tissue microarchitecture. As mentioned in the Introduction and shown later in the Results section, the broad bandwidth of the fiber optic hydrophone facilitates the resolution of both cancer cell clusters and the epithelium of prostate glands with smaller dimensions. The two components in combination form nonlinear spectral shapes. In this case, our previously used linear spectral analysis approach is no longer reliable. To address this issue, two nonlinear signal analysis methods were utilized in this study for quantifying the microarchitectural information encoded in PA signal. A PA signal from a phantom containing 100 µm diameter microspheres at the concentration of 20 per mm3, as shown in Fig. 4(a), was utilized as an example to explain the two signal analysis methods as described below.
Fig. 4. Process for quantitative PA analysis of tissue microstructure. (a) Representative PA signal measured by the all-optical needle PA sensing probe from a phantom containing 100 µm diameter microspheres at the concentration of 20 per mm3. (b) Power density spectra of the signal in (a) and the illustration of the expanding window for calculating spectral slopes for different spectral segments. (c) Calculated spectral slope values as a function of the upper limit of the expanding window. A linear fitting of this curve leads to the nonlinear spectral slope (i.e., tan θ, where θ is the angle between the fitting liner and x-axis). (d) The histogram of the signal in (a). Red line is the fitted generalized gamma distribution.
2.5.1 Nonlinear spectral analysis
After a radio-frequency (RF) PA signal of a sample was acquired, as shown in Fig. 4(a), the power density spectrum of the signal was generated, as shown in Fig. 4(b). Instead of performing a linear fit over the entire detected power spectrum, we fitted each segment of the power spectrum covered by an expanding window to a linear model. We defined the initial segment of the power spectrum covered by the expanding window as [3, 11 MHz]. 3 MHz was selected because the frequency response of the fiber optic hydrophone falls by -3dB, i.e. 70% of the maximum frequency response, below 3 MHz as shown in Fig. 1(e). 11 MHz was selected because most of power spectrum in this study reached the highest magnitude at this frequency. The linear fitting of the power spectrum in this segment led to a spectral slope. Then we increased the upper limit of the expanding window with a constant step size of 2 MHz. For each segment, a linear fitting of the power spectrum was conducted, leading to a spectral slope. This process continued until the expanding window reached [3, 29 MHz], where 29 MHz was the upper limit of the calibrated spectral range of the fiber optic hydrophone. Fig. 4(c) shows the spectral slope values calculated for each power spectrum segment as a function of the limits of the expanding window. Then, the curve in Fig. 4(c) was fitted by a linear model, and the slope of this linear model was quantified as the nonlinear spectral slope.
Compared to linear spectral slope, nonlinear spectral slope magnifies the contribution of high frequency components in quantifying the shape of a signal power spectrum. During the quantification of linear spectral slope, the low frequency components dominate the linear fitting due to their large magnitudes. By calculating the linear slopes within segments of a signal power spectrum in Fig. 4(b), the contribution of low frequency components is separately calculated. Any spectral peak in the high frequency range will lead to a local slow descending rate in the curve shown in Fig. 4(c), leading to the increase of the slope, i.e. tanθ, derived from the curve. Therefore, larger linear slopes derived from wide frequency windows in Figs. 4(b) and 4(c) indicate more high frequency components in the signal power spectrum and lead to higher nonlinear spectral slope values.
2.5.2 Signal envelope statistics
Besides the nonlinear spectral analysis in frequency domain, we also implemented envelope statistics to the PA signal in time domain, following a previous study [21]. Our study used the built-in “envelope” function in MATLAB (R2018b, Mathworks Inc., Natick, MA) to extract the positive envelopes of the signal by a Hilbert finite impulse response filter. The histogram of the signal envelopes was then generated using MATLAB, as shown in Fig. 4(d). The signal histogram was afterwards fitted to the probability density function of the Generalized Gamma (GG) distribution
where a, d and p are three positive parameters determining the shape of the distribution. The GG distribution has been used in PA signal analysis for envelope statistical studies [26,27]. Among them, the GG parameter a dominantly determines the height and the width of the distribution that are correlated to the microarchitectural information encoded in the RF PA and US signals [21]. Thus, GG parameter a was investigated in this study.2.5.3 Statistics and multivariate analysis
The correlations between the quantitative PA measurements, and the microsphere dimensions and concentrations in the phantom study were calculated. To examine whether the quantitative PA measurements acquired by the all-optical needle PA sensing probe can characterize the microarchitectures in prostate tissues, the nonlinear spectral slope, and the GG parameter a from different tissue categories (benign, GS=6, and GS=7) were compared in t-tests. T-tests were used because this study has limited sample size. The data sets were analyzed using the built-in function, “ttest”, in MATLAB (R2018b, Mathworks Inc., Natick, MA). The null hypothesis is defined as that the nonlinear spectral slope (or the GG parameter a) cannot differentiate among the three categories of prostate tissues. The correlations between the nonlinear spectral slope and the GG parameter a were also calculated to understand the dependence between the two parameters.
We further examined the feasibility of these quantitative PA measurements in distinguishing the phantoms and the prostate tissues via a multivariate support vector machine (SVM) analysis. SVM was conducted using a built-in function in MATLAB (“fitcsvm” function with SVMtrain, default Gaussian kernel with 2 box-constraint for higher cost of misclassified points) between each pair of the categories. Due to the limited sample size, we only attempted to define the boundaries among the tissue categories. The SVM boundaries were derived using the quantified PA measurements at each of the three laser wavelengths and all wavelengths. Then the accuracy of the boundaries in distinguishing the tissue categories was calculated.
3. Results
3.1 Spatial distribution of measurement sensitivity
Fig. 5 shows the spatial distribution of the SNR measured with the setup in Fig. 2(a) as an indication of the spatially distributed sensitivity enabled by the all-optical needle PA sensing probe. As shown in Fig. 5, measurements at the three wavelengths have similar sensitivity distribution patterns. Using 5 dB SNR as the detection threshold of the needle probe, the detection range of the needle PA probe is 2.83 ± 0.30 mm in x dimension (with the largest difference between the results at 1220 nm and 1370 nm), and 12.41 ± 0.43 mm in z dimension (with the largest difference between the results at 1220 nm and 266 nm.
Fig. 5. Spatial distributions of the detection SNR achieved by the all-optical needle PA sensing probe. (a), (b), and (c) are the results measured at 1220 nm, 1370 nm, and 266 nm, respectively. White lines are the 5 dB boundary at each wavelength.
3.2 Phantom study results
Fig. 6 shows the representative PA signals from phantoms containing optically absorbing microspheres with varied dimensions and concentrations, as well as the corresponding spectral analysis and envelope statistics. The RF PA signals in Fig. 6(a) show that, at the same concentration, microspheres with larger dimension generate higher PA signal amplitudes. In the power spectra in Fig. 6(b), compared to the 100 µm microspheres, the power density spectrum of 40 µm microspheres shows less steep descent along the frequency axis, indicating relatively more high frequency components compared to low frequency components. As demonstrated in Fig. 6(c), the quantified nonlinear spectral slope derived from the 40 µm microspheres is 0.01 which is higher than the -0.16 derived from the 100 µm microspheres. The RF PA signals in Fig. 6(e) show that, with the same dimension, the microspheres with higher concentration also generate higher PA signal amplitudes. The corresponding power density spectra of the two phantoms are shown in Fig. 6(f), where we can see that, despite the difference in magnitude, the overall shapes of the two spectra are similar. The difference between the quantified nonlinear spectral slopes derived from these phantoms containing microspheres with the same dimension but different concentrations (-0.08 vs. -0.19), as shown in Fig. 6(g), is not as significant as those in Fig. 6(e). Such a finding indicates that the nonlinear spectral slope is dominantly determined by the microsphere dimensions. The moderate difference (i.e., slightly smaller slope for the microspheres with higher concentration) can be due to the aggregation of the microspheres which leads to larger absorber dimensions and, therefore, stronger low frequency components in PA signal. In Figs. 6(d) and 6(h), microspheres with smaller dimensions and lower concentrations both had taller and narrower GG distributions, which agrees with the findings of a previous study [21]. Fig. 7 presents the statistical analysis of the nonlinear spectral slopes and the GG parameter a acquired from the phantoms containing microspheres with varied dimensions and concentrations. Figs. 7(a) and 7(c) are the quantified results from the phantoms containing microspheres with the same concentration (2 microspheres per mm3) but different dimensions (40, 50, 100, and 200 µm diameter); while Figs. 7(b) and 7(d) are the quantified results from the phantoms containing microspheres with the same dimension (100 µm diameter) but different concentrations (2, 6, 10, 12, and 20 microspheres per mm3). The nonlinear spectral slope shows a strong and negative correlation with the dimensions of the microspheres (r=-0.80, n=24, p<0.0001) but a relatively weak correlation with the concentrations of the microspheres (r=-0.54, n=30, p<0.0001); while the GG parameter a shows a strong correlation with the dimensions of microspheres (r=0.72, n=24, p<0.0001) and a moderate correlation with the microsphere concentrations (r=0.63, n=30, p<0.0001).
Fig. 6. Representative RF PA signals obtained for microsphere phantoms and the quantitative processing procedures including deriving the nonlinear spectral slopes and envelope statistics. The blue and orange curves in (a-d) are from microspheres at the same concentration of 2 per mm3, but with different dimensions of 40 µm and 200 µm, respectively. The gray and green curves in (e-h) are from microspheres with the same dimension of 100 -µm, but with different concentrations of 2 per mm3 and 20 per mm3, respectively. (a) and (e) RF PA signals. (b) and (f) The corresponding power spectra. (c) and (g) The derived spectral slopes using the expanding window and the quantification of the nonlinear spectral slopes. (d) and (h) The GG distributions and the corresponding parameter a calculated in envelope statistics. GG-a stands for GG parameter a.
Fig. 7. Statistical analysis of the quantitative PA measurements. (a) Nonlinear spectral slopes and (c) GG parameter a quantified from PA measurements of microspheres with the same concentrations of 2 counts per mm3 but varied dimesons (40, 50, 100, and 200 µm diameter). (b) Nonlinear spectral slopes and (d) GG parameter a quantified from PA measurements of microspheres with the same dimension (100 µm diameter) but varied concentrations (2, 6, 10, 12, and 20 microspheres per mm3).
Combining the nonlinear slopes and GG parameter a, Fig. 8 shows the SVM defining the boundaries among the phantoms with varied microsphere size and concentration. The accuracy as the ratio between the correctly categorized sample number over the total sample number is 92% in Fig. 8(a) and 80% in Fig. 8(b).
Fig. 8. SVM based categorization of the phantom study combining the nonlinear spectral slope and GG parameter a. (a) The classification between microsphere phantoms with the same concentrations of 2 counts per mm3 but varied dimesons (40, 50, 100, and 200 µm diameter). (b) The classification between microsphere phantoms with the same dimension (100 µm diameter) but varied concentrations (2, 6, 10, 12, and 20 microspheres per mm3).
3.3 Tissue study results
Fig. 9 shows the representative RF PA signals from the human prostate tissue samples at 1220 nm laser wavelength acquired by the all-optical needle PA sensing probe. The signals have an effective length of approximately 4 µs, corresponding to 60 wavelengths at the central frequency of the detection. There is no noticeable difference between the RF PA signals from the tissues in different categories (i.e. benign, GS=6, and GS=7).
Fig. 9. Representative RF PA signals acquired from a benign prostate tissue, a nonaggressive cancer tissue (GS=6), and an aggressive cancer tissue (GS=7), acquired at 1220 nm laser wavelength.
Fig. 10 shows the procedures for analyzing the PA signals from the prostate tissues. Fig. 10(a) demonstrates that, especially at 266 nm, the power spectra derived from the tissue samples shows significant nonlinearity. In the 20-30 MHz range, benign tissue possesses larger high frequency components than nonaggressive cancer (GS=6) which demonstrates larger high frequency components than aggressive cancer (GS=7). In Fig. 10(b), the curves derived from the expanding frequency window show more distinguishable trends among the tissue samples compared to the power spectra in Fig. 10(a). The nonlinear spectral slopes are in the order of benign, GS=6 and GS=7, from the largest to smallest at all three wavelengths. This observation is supported by the histology in Fig. 3. As prostate cancer progresses, the simple layer of epithelial cells in benign tissue or nonaggressive cancer that produces high frequency signal components may be replaced by large architecturally complex cell clusters in aggressive cancer. The large gaps between the cancer cell clusters, i.e. connective tissue rich in lipid and collagen, also tend to produce less high frequency spectral components at 1220 nm and 1370 nm compared to the benign tissue. These observations also match with the findings in phantom studies, where the microspheres dimensions are negatively correlated with the nonlinear spectral slopes.
Fig. 10. Nonlinear spectral analysis and envelope statistics of representative PA signals acquired from human prostate tissues, including benign (green curves), GS=6 (blue curves), and GS=7 (orange curves), at the three optical wavelengths (1220 nm, 1370 nm, and 266 nm). (a) Power spectra of the PA signals acquired from the three prostate tissues at the three optical wavelengths. (b) Nonlinear spectral slopes calculated from the power spectra in (a). (c) GG distributions derived from the envelope statistics of PA signals acquired from the three prostate tissues at the three optical wavelengths. GG-a stands for GG parameter a.
In Fig. 10(c), GG distributions show increasingly narrow and tall profiles along with the progression of prostate cancer at all three wavelengths. The three tissue groups show more observable differences at 1220 nm and 1370 nm compared to those at 266 nm. Such results are supported by the fact that the connective tissue rich in lipid and collagen contents decreases as the cancer progresses [9,28,29]. The GG distribution at 266 nm in Fig. 10(c) shows less differentiable features, which is supported by the fact that the diagnosis of aggressive prostate cancer frequently involves architecture changes instead of changes in the number of cancer cells within the tissues. The monotonically decreasing trend of quantified GG parameter a at 1220 nm and 1370 nm (targeting lipid and collagen, respectively) also matches the findings in phantom study that the GG parameter a is positively correlated with the concentrations of the optical absorbers.
With the nonlinear spectral analysis and envelope statistics results from all the prostate tissues, including 14 benign tissues, 8 nonaggressive cancers (GS=6), and 8 aggressive cancers (GS=7), statistical analysis were conducted, as shown in Fig. 11. At each of the three wavelengths (i.e., 1220 nm, 1370 nm, and 266 nm), the quantified nonlinear spectral slope can differentiate between any two of the three categories (i.e., benign, GS=6, and GS=7) with statistical significance (p<0.05). The quantified GG parameter a at the 1220 nm and 1370 nm wavelengths can also differentiate between any two of the three categories with statistical significance (p<0.05). The quantified GG parameter a at the 266 nm wavelength, however, cannot differentiate between the benign and the GS=6 categories (p>0.05), although the GS=6 and the GS=7 categories can be differentiated (p<0.05). These results are clinically desirable, as over-diagnosis of GS 6 prostate cancer leads to overtreatment and morbidity [30,31].
Fig. 11. Statistical analyses of the nonlinear spectral slopes and the GG parameter a acquired from the three categories of human prostate tissues (benign, GS=6, and GS=7). (a-c) are box plots showing the quantified nonlinear spectral slopes of the three categories of tissues measured at the wavelength of 1220 nm, 1370 nm, and 266 nm, respectively. (d-f) are box plots showing the GG parameter a of the three categories of tissues measured at the wavelength of 1220 nm, 1370 nm and 266 nm, respectively. p values are quantified for the t-tests comparing the results between any two of the three categories.
As shown in Table 1, the quantified correlations between the nonlinear spectral slope and the GG parameter a at all the wavelengths are no larger than 0.6, suggesting that the nonlinear spectral slope and the GG parameter a contain independent diagnostic information. This justifies the use of multivariate analysis involving these two different quantitative PA parameters.
Table 1. Correlation between quantitative measurements and SVM results
As shown in Fig. 12, the SVM at any of the three optical wavelengths defined the boundaries between any two of the three tissue categories. We calculated the accuracy of the boundaries as the ratio between the correctly categorized sample number over the total sample number. Table 1 lists the accuracy of the boundaries defined using either of the two quantitative parameters (i.e., nonlinear slope and GG parameter a). When combining the quantitative parameters at all the three wavelengths, SVM demonstrated 100% accuracy in defining the boundaries among the three tissue categories.
Fig. 12. SVM based categorization of the human prostate tissues, including benign, GS=6, and GS=7, by combining the nonlinear spectral slope and GG parameter a quantified at each of the optical wavelengths (1220 nm, 1370 nm, and 266 nm).
4. Conclusion and discussion
This study, for the first time, validated the feasibility of an all-optical needle PA sensing probe for interstitial measurements and characterization of biological tissues. With the updated signal analysis methods, quantitative PA measurements from this all-optical needle PA sensing probe demonstrated the capability in quantifying the dimensions and concentrations of the optical absorbers in phantoms, and in differentiating among ex vivo human prostate tissues with different stages of progressive cancer.
The main purpose of the phantom study is to examine the performance of the measurement setup with simple and well controlled conditions. Therefore, the experiments were restricted to microspheres with uniform dimensions and distributions. The nonlinear slope and the GG parameter of 200-µm microspheres in Fig. 7 showed larger error bar, probably due to the aggregation of the microspheres. Thus, we did not use larger microspheres. We do not expect to differentiate the concentrations of the microspheres using nonlinear slope, as the shape of the signal power spectrum is mostly related to the microsphere dimensions. The overlapping between GG parameters of the microsphere at varied concentrations may indicate that the concentration difference investigated in this study is close to the sensitivity limit of magnitude-based envelope statistics.
The nonlinear spectral slope uses an expanding window to observe both 1) the low frequency components separately from the high frequency ones, and 2) the relative magnitude of low and high frequency components. Therefore, the expanding window bandwidth and step size have to fit to the specific microarchitectural dimensions in the targeted samples. Such approach assisted in comprehensively quantifying shape of a PA signal power spectrum that encodes tissue microarchitecture at multiple scales. The envelope statistics has shown the ability to quantify additional tissue component information (i.e., changes in lipid and collagen contents). Most parameters at the studied wavelengths can differentiate cancerous tissues from benign tissues and differentiate aggressive cancerous tissues (GS=7) from nonaggressive ones (GS=6). These parameters may be able to differentiate between clinically significant prostate cancer and benign tissue or clinically insignificant prostate cancer.
Considering that the nonlinear spectral slope and the GG parameter a contain independent diagnostic information, multivariate analysis combines the diagnostic information from these two parameters and takes full advantage of the all-optical PA measurement approach to improve diagnostic accuracy. As shown in this study, measurable SVM boundaries were identified between the phantoms and the three types of tissues. Future study with more tissue samples will validate these boundaries [9] and investigate the mapping of these quantitative parameters to individual Gleason grades.
The needle probe presented has demonstrated the effectiveness in acquiring PA signals with sufficient temporal length for statistical analysis in both frequency domain and time domain.
The sensitivity profiles of the needle PA sensing probe in Fig. 5 were determined by both the optical transmission and the acoustic reception. The profiles were measured in an ideal situation where a small interpolation occurred in a homogeneous background. The profiles may be slightly deformed by the heterogeneous optical and acoustic properties in tissue. The sensitivity profiles were measured only in the plane containing both the transmission and reception components, and on the side where the fiber optic hydrophone is located. Owing to the uniform optical transmission in the azimuthal dimension of the fiber optic diffuser, the sensitivity profiles in other planes are similar except on the opposing side of the fiber optic diffuser to the hydrophone, where the aperture of the hydrophone is blocked. We assume the worst case that half of the reception angle of the fiber optic hydrophone is blocked. Therefore, we estimated sampling volume of the needle PA probe by approximating the SNR distributions in Fig. 5 as triangles and rotating the triangle along the z axis by 180 degrees. The sampling volume covered by the needle PA probe is 52 mm3 (=πr2h/3/2=π×2.832×12.41/3/2). On the other hand, considering that the sampling cavity of an 18 Gauge biopsy needle has the shape of a half cylinder with diameter of 0.838 mm and height of half an inch, the sampling volume of a biopsy core is only approximately 3.5 mm3 (=πr2h/2= π×0.4192×12.7/2). Therefore, the needle PA probe has the potential to significantly increase the sampling volume for prostate cancer diagnosis. When the sampling volume covers both benign and cancerous tissue, the quantitative parameters may be averaged between the two categories. However, the detectable volume of the needle PA probe is much smaller than a clinically significant prostate tumor (10 mm in diameter [19]). Therefore, multiple insertions of the needle probe with the proposed 5 mm separation should be able to sample within the tumor volume where only cancerous tissue exist.
To prepare the all-optical needle PA sensing probe for clinical translation in the future, we will integrate the fiber optic diffuser and the fiber optic hydrophone into a steel needle with side window. The internal surface of the needle will be polished or coated for good light reflectance, which will increase the optical energy delivery efficiency and help reduce the optical energy at 266 nm below ANSI safety limit. The exposed optical components will be protected by a layer of medical grade polyurethane, which is acoustically matched with biological tissue and with negligible optical attenuation. We will also investigate on improving the sensitivity of the fiber optic hydrophone using established methods [32] to reduce the optical density at 266 nm while maintaining the SNR profiles achieved in Fig. 5. Optical wavelengths between 750 to 850 nm targeting hemoglobin content will be included for measurements in vivo. In addition, we will search for an optimal design for light illumination by adjusting the diameter and the emission length of the fiber optic diffuser. A larger fiber diameter allows for more optical energy coupling, but the diameter is restricted by the inner diameter of the steel needle. Extended emission length facilitates more reliable statistics-based signal analysis; however, the corresponding increase in emission surface area will reduce the deliverable optical density as well as the SNR. All these optimization processes will be implemented in the next step of this research.
Funding
National Eye Institute (P30 EY007003); National Cancer Institute (1R01CA25049901A1); Prostate Cancer Foundation (19YOUN12); National Institute of Diabetes and Digestive and Kidney Diseases (1R01DK12568701); National Cancer Institute (5R37CA22282903).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. V. Kumar, A. Abbas, and J. Aster, Robbins & Cotran Pathologic Basis of Disease (Elsevier, 2009).
2. D. Joy, V. R. Thava, and B. B. Scott, “Diagnosis of fatty liver disease: is biopsy necessary?” Eur. J. Gastroenterol. Hepatol. 15(5), 539–543 (2003). [CrossRef]
3. L. Gortzak-Uzan, W. Jimenez, S. Nofech-Mozes, N. Ismiil, M. A. Khalifa, V. Dubé, B. Rosen, J. Murphy, S. Laframboise, and A. Covens, “Sentinel lymph node biopsy vs. pelvic lymphadenectomy in early stage cervical cancer: Is it time to change the gold standard?” Gynecol. Oncol. 116(1), 28–32 (2010). [CrossRef]
4. S. Loeb, S. van den Heuvel, X. Zhu, C. H. Bangma, F. H. Schröder, and M. J. Roobol, “Infectious complications and hospital admissions after prostate biopsy in a european randomized trial,” Eur. Urol. 61(6), 1110–1114 (2012). [CrossRef]
5. G. Xu, I. A. Dar, C. Tao, X. Liu, C. X. Deng, and X. Wang, “Photoacoustic spectrum analysis for microstructure characterization in biological tissue: a feasibility study,” Appl. Phys. Lett. 101(22), 221102 (2012). [CrossRef]
6. G. Xu, J. B. Fowlkes, C. Tao, X. Liu, and X. Wang, “Photoacoustic Spectrum analysis for microstructure characterization in biological tissue: analytical model,” Ultrasound Med. Biol. 41(5), 1473–1480 (2015). [CrossRef]
7. G. Xu, Z. Meng, J. Lin, C. X. Deng, P. Carson, J. Fowlkes, C. Tao, X. Liu, and X. Wang, “High resolution Physio-chemical Tissue Analysis: Towards Non-invasive In Vivo Biopsy,” Sci. Rep. 6(1), 16937 (2016). [CrossRef]
8. G. Xu, Z.-X. Meng, J. D. Lin, J. Yuan, P. L. Carson, B. Joshi, and X. Wang, “The functional pitch of an organ: quantification of tissue texture with photoacoustic spectrum analysis,” Radiology 271(1), 248–254 (2014). [CrossRef]
9. S. Huang, Y. Qin, Y. Chen, J. Pan, C. Xu, D. Wu, W.-Y. Chao, J. T. Wei, S. A. Tomlins, X. Wang, J. B. Fowlkes, P. L. Carson, Q. Cheng, and G. Xu, “Interstitial assessment of aggressive prostate cancer by physio-chemical photoacoustics: An ex vivo study with intact human prostates,” J. Med. Phys. 45(9), 4125–4132 (2018). [CrossRef]
10. G. Xu, M. C. Davis, J. Siddiqui, S. A. Tomlins, S. Huang, L. P. Kunju, J. T. Wei, and X. Wang, “Quantifying Gleason scores with photoacoustic spectral analysis: feasibility study with human tissues,” Biomed. Opt. Express 6(12), 4781–4789 (2015). [CrossRef]
11. E. Amidi, A. Mostafa, S. Nandy, G. Yang, W. Middleton, C. Siegel, and Q. Zhu, “Classification of human ovarian cancer using functional, spectral, and imaging features obtained from in vivo photoacoustic imaging,” Biomed. Opt. Express 10(5), 2303–2317 (2019). [CrossRef]
12. P. J. van den Berg, R. Bansal, K. Daoudi, W. Steenbergen, and J. Prakash, “Preclinical detection of liver fibrosis using dual-modality photoacoustic/ultrasound system,” Biomed. Opt. Express 7(12), 5081–5091 (2016). [CrossRef]
13. G. Xu, Y. Xue, Z. G. Özkurt, N. Slimani, Z. Hu, X. Wang, K. Xia, T. Ma, Q. Zhou, and H. Demirci, “Photoacoustic imaging features of intraocular tumors: Retinoblastoma and uveal melanoma,” PLoS One 12(2), e0170752 (2017). [CrossRef]
14. E. Hysi, R. K. Saha, and M. C. Kolios, “Photoacoustic ultrasound spectroscopy for assessing red blood cell aggregation and oxygenation,” J. Biomed. Opt. 17(12), 125006 (2012). [CrossRef]
15. P. V. Chitnis, J. Mamou, and E. J. Feleppa, “Spectrum analysis of photoacoustic signals for characterizing lymph nodes,” J. Acoust. Soc. Am. 135(4), 2372 (2014). [CrossRef]
16. T. Feng, Q. Li, C. Zhang, G. Xu, L. J. Guo, J. Yuan, and X. Wang, “Characterizing cellular morphology by photoacoustic spectrum analysis with an ultra-broadband optical ultrasonic detector,” Opt. Express 24(17), 19853–19862 (2016). [CrossRef]
17. S. W. Flax, N. J. Pelc, G. H. Glover, F. D. Gutmann, and M. McLachlan, “Spectral characterization and attenuation measurements in ultrasound,” Ultrason. Imaging 5(2), 95–116 (1983). [CrossRef]
18. H. Zhang, W. Chao, Q. Cheng, S. Huang, X. Wang, D. Wu, and G. Xu, “Interstitial photoacoustic spectral analysis: instrumentation and validation,” Biomed. Opt. Express 8(3), 1689–1697 (2017). [CrossRef]
19. T. A. Stamey, F. Freiha, J. McNeal, E. Redwine, A. Whittemore, and H. Schmid, “Localized prostate cancer. Relationship of tumor volume to clinical significance for treatment of prostate cancer,” Cancer 71(3), 933–938 (1993). [CrossRef]
20. J. Jo, J. Siddiqui, Y. Zhu, L. Ni, S. R. Kothapalli, S. A. Tomlins, J. T. Wei, E. T. Keller, A. M. Udager, X. Wang, and G. Xu, “Photoacoustic spectral analysis at ultraviolet wavelengths for characterizing the Gleason grades of prostate cancer,” Opt. Lett. 45(21), 6042–6045 (2020). [CrossRef]
21. E. Hysi, M. N. Fadhel, M. J. Moore, J. Zalev, E. M. Strohm, and M. C. Kolios, “Insights into photoacoustic speckle and applications in tumor characterization,” Photoacoustics 14, 37–48 (2019). [CrossRef]
22. W.-H. Choi and I. Papautsky, “Fabrication of a needle-type pH sensor by selective electrodeposition,” J. Micro/Nanolith. MEMS MOEMS 10(2), 020501 (2011). [CrossRef]
23. P. Morris, A. Hurrell, A. Shaw, E. Zhang, and P. Beard, “A Fabry–Pérot fiber-optic ultrasonic hydrophone for the simultaneous measurement of temperature and acoustic pressure,” J. Acoust. Soc. Am. 125(6), 3611–3622 (2009). [CrossRef]
24. P. Lai, X. Xu, and L. V. Wang, “Dependence of optical scattering from Intralipid in gelatin-gel based tissue-mimicking phantoms on mixing temperature and time,” J. Biomed. Opt. 19(3), 035002 (2014). [CrossRef]
25. S. T. Flock, S. L. Jacques, B. C. Wilson, W. M. Star, and M. J. C. V. Gemert, “Optical properties of intralipid: a phantom medium for light propagation studies,” Lasers Surg. Med. 12(5), 510–519 (1992). [CrossRef]
26. E. W. Stacy, “A Generalization of the Gamma Distribution,” Ann. Math. Stat. 33(3), 1187–1192 (1962). [CrossRef]
27. O. Gomes, C. Combes, and A. Dussauchoy, “Parameter estimation of the generalized gamma distribution,” Math. Comput. Simul. 79(4), 955–963 (2008). [CrossRef]
28. Y. Pu, W. Wang, G. Tang, and R. R. Alfano, “Changes of collagen and nicotinamide adenine dinucleotide in human cancerous and normal prostate tissues studied using native fluorescence spectroscopy with selective excitation wavelength,” J. Biomed. Opt. 15(4), 047008 (2010). [CrossRef]
29. C. Morrison, J. Thornhill, and E. Gaffney, “The connective tissue framework in the normal prostate, B.P.H and prostate cancer: analysis by scanning electron microscopy after cellular digestion,” Urol. Res. 28(5), 304–307 (2000). [CrossRef]
30. D. C. Grossman, S. J. Curry, D. K. Owens, K. Bibbins-Domingo, A. B. Caughey, K. W. Davidson, C. A. Doubeni, M. Ebell, J. W. E. Jr, A. R. Kemper, A. H. Krist, M. Kubik, C. S. Landefeld, C. M. Mangione, M. Silverstein, M. A. Simon, A. L. Siu, and C.-W. Tseng, “Screening for prostate cancer: us preventive services task force recommendation statement,” JAMA 319(18), 1901–1913 (2018). [CrossRef]
31. R. Gulati, L. Y. T. Inoue, J. L. Gore, J. Katcher, and R. Etzioni, “Individualized estimates of overdiagnosis in screen-detected prostate cancer,” J. Natl. Cancer Inst. 106(2), djt367 (2014). [CrossRef]
32. J. A. Guggenheim, J. Li, T. J. Allen, R. J. Colchester, S. Noimark, O. Ogunlade, I. P. Parkin, I. Papakonstantinou, A. E. Desjardins, E. Z. Zhang, and P. C. Beard, “Ultrasensitive plano-concave optical microresonators for ultrasound sensing,” Nat. Photonics 11(11), 714–719 (2017). [CrossRef]
