1. Introduction

During the past decades, prostate cancer (PCa), with an annual incident rate much higher than any other cancer, is the most commonly diagnosed cancer in American men [1]. PCa has a relatively low progression rate. Patients with early diagnosed PCa have a five year survival rate close to 100%, yet the percentage decreases dramatically once the cancer has metastasized [1]. Identifying aggressive from indolent PCa to prevent metastasis and death is critical to improving outcomes for patients with PCa.

Although serum prostate specific antigen (PSA) is widely used for PCa early detection, PSA screening results in over-diagnosis and over-treatment with unclear mortality benefit [2–18]. Likewise, sensitivity and specificity limitations of serum PSA are well documented. Serum PSA is strongly correlated to prostate volume, and can be elevated in many benign conditions and increases with normal aging [19–21]. Several serum and urine based biomarkers (e.g. 4Kscore, phi, PCA3, MiPS) have been advanced that improve upon the PSA for predicting the presence of PCa on biopsy [19, 22–24]. Unfortunately, none of these markers are specific for high grade aggressive PCa. Elevated serum PSA, along with results from these tests or physical examination (e.g. digital rectal exam), will lead to ultrasound (US) guided, transrectal US (TRUS) biopsy, the standard procedure for evaluating the presence and aggressiveness of PCa. The microscopic architecture of the biopsied tissues, stained and visualized by standard hematoxylin and eosin (H&E) or immunohistochemical (IHC) histology, are evaluated by pathologists. PCa is assigned a Gleason score [25], a highly prognostic architectural based grading system for PCa. Each Gleason score is the summation of primary and secondary Gleason grades. The 5 Gleason grades are illustrated in Fig. 1(a) [25]. Gleason scores thereby vary between 2 to 10. As US imaging has limited sensitivity to PCa, transrectal biopsies are typically performed following a predetermined pattern overlaid onto the prostate contour delineated by US, yielding 20-30% false negative rates [26–28]. Transperineal saturated biopsy samples more than 50 sites in a prostate [29]. Less than 10% of the sample cores are clinically significant yet the false negative rate could still be as high as 20% at initial biopsy [28]. Repeated biopsies are performed in patients with negative initial biopsies yet continuously increased PSA levels, leading to added diagnostic costs, anxiety, pain, potential infection, and unrecoverable damage to the neurovascular bundles and consequently erectile dysfunction [30].

 

Fig. 1 Simulation of PASA on Gleason patterns. (a) The Gleason patterns. (b) Representative simulated signals generated by Gleason grades 1 and 5. The fine architecture of Gleason grade 1 is below the ultrasonic resolution (200 µm, determined by the default sensor bandwidth of 0-15MHz). The signal form Gleason grade 1 is smooth and thereby carries less high frequency components than that generated by Gleason grade 5. (c) The power spectra of the PA signals generated by Gleason grades 1-5. The linear fit to the power spectrum for Gleason grade 1 was plotted in red dashed line, of which the slope is tan(θ).

Download Full Size | PDF

Alternative imaging modalities for estimating Gleason grading to guide biopsies or provide independent noninvasive diagnosis has been attempted. Prebiopsy magnetic resonance imaging (MRI) has been recently introduced to TRUS biopsy for providing suspicious malignant regions within the prostate. However, MRI contrast is based on neovasculature or microcirculation which is only significant in advanced PCa. Moreover, evaluation of MRI results requires experienced radiologists. 10-20% false negative MRI-TRUS guided biopsy cases are reported, especially for those with low but clinically significant Gleason scores [31–33]. Conventional optical spectroscopy has demonstrated correlation to the Gleason scores by quantifying histochemical components in the cancerous tissue by optical absorption and fluoresence [34–36]. In addition, optical backscattering has also been investigated to quantify the dimensions of the microscopic architecture of prostate [34] and other cancerous tissue at micron to submicron dimensions [37]. Nonetheless, optical spectroscopy has limited capability in assessing deep tissue volumes non-invasively due to the overwhelming optical scattering in biological tissues. Recently, frequency domain analysis of US signals, namely quantitative US (QUS), has demonstrated the capability of differentiating benign and cancerous prostate tissue in vivo [38–40]. QUS, purely based on the tissue ultrasonic properties, is not capable of selectively probing the chemical and molecular information in tissue.

Photoacoustic (PA) imaging and sensing is a non-ionizing and non-radioactive modality combining the advantages of optical and US imaging [41–43]. Frequency domain analysis of PA measurements, namely PA spectral analysis (PASA) [44, 45], has demonstrated the potential of assessing microscopic features in biological tissue [46–50]. PASA follows the framework of QUS, as described in detail in our previous publications [44, 45]. Taking advantage of the unique optical selectivity of PA measurements, PASA can identify the histological distributions of individual chemical components at ultrasonic resolution. Our recent study on PASA has successfully characterized the histological architecture of lipid clusters in the fatty livers [48]. Other investigations of PASA include the characterization of the disease conditions in lymph nodes [49] and the morphology of red blood cells [50]. PASA has also been attempted in PCa diagnosis [46, 47]. However, the investigations are limited to the increase of neoangiogenesis within the tumors.

In this study, PASA was implemented to directly assessing the Gleason patterns. PASA was simulated with the classic illustration of the Gleason patterns in [25]. H&E stained ex vivo human prostate tissues were scanned to further test the feasibility of this approach. The results from this study suggest that PASA could provide diagnosis of PCa by quantifying the microscopic architectures in the cancerous tissues.

2. Method and material

2.1 Simulations on classic Gleason patterns

The classic Gleason patterns in reference [25] is reproduced in Fig. 1(a). In Fig. 1(a) [25], the dark pixels, representing the cancer cells, formulate a series of cluster patterns. The heterogeneity of the cluster patterns increases along with the cancer progression. Gleason scores are formulated by adding the primary and secondary Gleason grades in a tissue sample. The PA signals generated by Gleason patterns were simulated by the MATLAB toolbox K-wave [51] with a sampling rate of 300MHz. The Gleason patterns were scaled to their approximate original dimensions in tissue, i.e. the total width of Fig. 1(a) was scaled to 450 µm. The dark pixels in Fig. 1(a) were assigned as the PA sources with initial acoustic pressure of 1. The background was assigned acoustic pressure 0. The speed of sound was set as 1510 m/s. 7 point acoustic sensors were used for each pattern. The sensors were evenly distributed at the bottom of each pattern in Fig. 1(a). The power spectra of the simulated PA signals were calculated with Welch approach [52] with pwelch function in MATLAB (R2011, MathWorks, Natick, MA). During the calculation, the power spectra were calculated within a sliding window of 1/4 of the total signal length and 60% overlap per sliding step. The power spectra derived from all sliding steps and all sensors for the same Gleason pattern were averaged to formulate smooth power spectra representing the microscopic architecture of each Gleason pattern. The frequency dependent attenuation was calculated by the method presented in our previous studies [44, 45]. As illustrated by the red dashed line in Fig. 1(b), the [0.1, 10] MHz range of the simulated power spectra were fitted to linear models. The slope of the power spectrum generated by Gleason pattern 1 was illustrated in Fig. 1(c). Higher slope values (i.e. negative slopes with smaller absolute values) indicate more high frequency components in the spectrum and thereby more heterogeneous tissue architectures. Since the areas of the Gleason patterns were different, the magnitudes of the power spectra were not considered in this study.

2.2 PASA of ex vivo human prostate tissues

Four types of human prostate tissues including benign (4 samples), Gleason 6 (3 + 3, 4 samples), Gleason 7 (3 + 4, 5 samples) and Gleason 9 (4 + 5, 6 samples) were used in this study. Two neighboring slices with thickness of 100 µm cut from the prostate sample were observed in pair. Both slices were fixed on glass slides and stained by standard H&E procedure. One slice was sealed by a cover glass and observed by a pathologist. The representative histology photographs are shown in Fig. 2(a)-2(d). The other slice was scanned immediately without cover glass using the setup illustrated in Fig. 3. As shown in Fig. 3(a), the sample slice was covered by a large volume of US gel. A needle hydrophone (HNC-1500, ONDA Co.) was inserted into the gel volume for PA signal reception. The hydrophone has nominal bandwidth of 0.1-10MHz but capable of covering frequency up to 20MHz. The optical illumination was generated by an Nd:YAG laser (PowerLite 8010, Continuum, CA) working at 532 nm. The laser beam was expanded to cover the whole sample area with an optical density of 6 mJ/cm². 532 nm was used due to the high optical absorption of the pink H&E stain at this wavelength. The PA signals acquired by the hydrophone were amplified by 40 dB and then digitized by an oscilloscope (TDS 540, Tektronix, Beaverton, OR) at a sampling frequency of 250 MHz. The power spectra of the PA signals were calculated and calibrated by subtracting the frequency response of the data acquisition system (provided by the hydrophone manufacturer) in logarithmic scale. The frequency response was linearly interpolated to the frequency resolution when calibrating the power spectra in the experiments. The frequency dependent attenuation to the PA signals was compensated following the method described in our previous publications [44, 45, 53]. As shown in Fig. 3(b), each sample was rotated and measured along four different orientations. The power spectra corresponding to the four orientations were averaged. The representative power spectra of the PA signals are shown in Fig. 2(g). The PASA slope of each sample was calculated.

 

Fig. 2 Histology photographs of prostate tissues used in this study and representative PA spectra generated by the tissues. (a) Normal. (b) Gleason 6. (c) Gleason 7. (d) Gleason 9. Arrows indicate the clustered cancer cells. Scale bar: 250 μm. (e) and (f) are representative PA signals generated by a benign and a Gleason 9 samples, respectively. More fluctuations can be seen in the PA signal from the Gleason 9 sample. (g) The power spectra of the PA signals generated by the prostate tissues with different Gleason scores. Black dashed line indicates the noise level. Due to the low signal-to-noise ratio (SNR) in the experiment case, only the spectra above the noise floor (black dashed line) were fit to linear models. The frequency range for linear fitting to the Gleason 9 spectrum in (g) is from 0.1 MHz to 3.8 MHz as marked by the cyan dashed line. The slope of the linear fit to Gleason 9 spectrum (magenta dashed line) is tan (θ).

Download Full Size | PDF

 

Fig. 3 (a) Setup for PA scan of human PCa tissues. The sliced prostate tissue was fix on a glass slide and stained by H&E staining. The glass slide was covered by US gel and attached to the surface of a tank of water. A needle hydrophone was inserted into the gel volume for PA wave reception. The PA signals were amplified, recorded by an oscilloscope and post-processed by a PC. 532-nm laser was used due to the high optical absorption of the pink H&E stain at this wavelength. (b) Each sample was measured along four different orientations.

Download Full Size | PDF

3. Results

3.1 Simulation results

Simulation results in Fig. 4 show that: 1) the PASA slopes increase along with the Gleason grades in a parabolic trend; and 2) the slope values increase approximately linearly with respect to the Gleason grades at higher grades (Gleason 3-5), as marked by the green dashed line in Fig. 4.

 

Fig. 4 Averaged slopes from PASA for each Gleason grade. At higher Gleason grades (i.e. 3-5), the slope values appear to increase linearly with respect to the Gleason grade, as marked by the green dashed line.

Download Full Size | PDF

3.2 Ex vivo tissue

In H&E stained histological photographs in Fig. 2, normal cells appear in light color whereas cancer cells appear in dark color. The cells in the prostate tissues become less differentiated as the Gleason scores increases. The dark cancer cells clustered together, increasing the heterogeneity of the tissue and leading to high frequency components in the PA signals. The slopes derived from the samples were shown in Fig. 5(a). Since the benign prostate tissue was not studied in simulation, boxplot in Fig. 5(b) only contains data acquired from the cancerous tissues [marked by green dashed box in Fig. 5(a)]. The horizontal axis in Fig. 5(b) is intentionally scaled linearly to show the linear correlation between the Gleason scores and the PASA slopes, as indicated by the green dashed line in Fig. 5(b). Considering the Gleason scores are linear combination of the Gleason grades, the linear increase trend of the slopes with respect to the Gleason scores also validated our finding in Fig. 4 where the slope is approximately linear correlated to Gleason grade of 3-5.

 

Fig. 5 Results from ex vivo human prostate tissues. (a) Slope values acquired from normal and cancerous prostate tissues. (b) Boxplot of the cancerous data with respect to Gleason scores.

Download Full Size | PDF

4. Discussions

The SNR in the ex vivo tissue experiment was slightly low due to the fact that the tissue was thinly sliced for H&E staining. The small tissue volume limited the signal magnitude against the background noise. The signals reflected by the glass slide could also be a factor reducing the SNR. The glass slide was seated in a water tank to couple part of the PA energy at the bottom surface of the glass slide and minimize reverberation. Figure 2(e) and 2(f) show that the signal reflection was not significant. In future study with tissue chunks, the samples will be submerged in water or sealed in gel phantom to avoid acoustic reflection.

The low SNR in the ex vivo tissue experiment leads to the variation of the PASA approach from that in our previous studies [44, 45]. Instead of getting the slope values within a fixed frequency range, the slope values in this work were calculated from the linear fit in the frequency range where the PA power spectra were above the system noise level, as indicated in Fig. 2(g). This slope calculation method is also the reason for the large variation of the slopes from the benign tissues. Because of the lack of high frequency spectral components, the PA spectra from benign tissues dropped quickly below noise level. In this case, the slopes were calculated by dividing the intercept of the linear fit by a small frequency width less than 1 MHz. The variation of the slopes thereby includes the magnified variations of the intercepts. The low SNR issue could be alleviated when taking measurements in a 3D tissue volume. Compared to the 2D pattern in the thinly sliced samples, a tissue volume including identical tissue architecture will absorb more optical energy and produce higher signal magnitudes over the noise level.

As shown in Figs. 4 and 5, the slopes in simulation and experiment have similar correlation trends to the Gleason grades/scores. The linear correlation between the PASA slopes and Gleason 7-9 are estimated based on the available data yet higher order model could be more accurate when measurements at more Gleason scores are acquired. However, the absolute values of the slopes from the simulations and experiments do not match with each other. The main reason for such mismatch could be that the Gleason patterns for simulation were illustrated in binary images, which led to sharp transitions between the cancer cells and the background. Such sharp transitions generated more high frequency components in the simulated PA signals. The contrast transitions in actual tissues were significantly smoother. The measured PA signals from ex vivo tissues thereby produced lower slope values. In addition, the inability to scale the Gleason patterns to the exact dimension in tissue also led to the shift of the spectral components along the frequency axis in the simulation. Another reason for the mismatch between the simulation and experiment result is that each ex vivo sample was a mix of the benign and cancerous tissue. The tissue heterogeneity averaged over the whole sample area could lead to lower slopes in the experiment results compared to those from simulation.

The Welch approach [52] produces smooth power spectra by averaging power spectra calculated within a slide window in time domain. The sliding window was defines as 1/4 of the total signal length to provide frequency resolution higher than that of the hydrophone frequency response provided the manufacturer (0.1MHz). The frequency resolution in simulations and experiments in this study were 0.006 MHz [ = (300MHz/2)/(105000 sampling points /4)] and 0.033 MHz [ = (250MHz/ 2)/(15000 sampling points /4)] frequency domain resolution in, respectively.

The ultimate implementation of the PASA methodologies in clinical diagnosis of PCa will be restricted by the acoustic and optical attenuation when assessing deep tissue. We thereby designed a fine needle based PASA probe for interstitial measurements. The needle probe will be described in an independent article. In brief, the needle probe has diameter smaller than a biopsy needle and can fit into the core of a biopsy needle. The needle probe can deliver a cylindrical light source and acquired PA signals close to the light source. Attenuation of high frequency component in the PA signals could be minimized, facilitating the observation of detailed microarchitecture in deep tissue.

5. Conclusion and future works

The simulation with the classic Gleason patterns and the experiment on human PCa tissues have demonstrated the feasibility of quantifying the Gleason grades/score with the recently developed PASA method. In the future, study involving larger number of human PCa samples at different Gleason grades will be performed to formulate a mapping curve that can map PASA slopes to Gleason scores. Before the PASA method can be translated to clinic, a mechanism for in vivo staining of PCa tumors with high sensitivity and high specificity needs to be established. As the Gleason patterns are composed by PCa cells, cancer cell targeting and optical contrast enhancing nanoparticles could be a promising choice [54]. Powered by the highly specific protein targeting delivery mechanism, the nanoparticles uptaken by PCa cells can facilitate high-contrast imaging of tumors and in vivo characterization of histological microstructures in the prostate. Once validated, PASA may benefit the clinical management of PCa by providing a minimally-invasive and powerful diagnostic tool which could be either an alternative or an addition to conventional needle biopsy and histopathology.