Prostate cancer is the second leading cause of cancer death in American men. Screening by PSA testing and digital rectal examination often leads to transrectal ultrasound (TRUS)-guided biopsy, but conventional ultrasound has poor sensitivity for detecting prostate cancer because roughly half of all prostate cancers are isoechoic - meaning they look the same as normal prostate tissue on ultrasound. Even systematic biopsy schemes that sample multiple regions carry a greater than 30% chance of misleading diagnosis from sampling error.
Attempts to improve the sensitivity of ultrasound-based prostate imaging - including contrast-enhanced ultrasound and real-time elastography - have not achieved adequate cancer detection rates. MRI provides excellent tissue characterization but is expensive, time-consuming, and typically reserved for staging advanced disease rather than initial screening. A new imaging approach is needed that combines the accessibility and real-time capability of ultrasound with better tissue differentiation.
Photoacoustic imaging is a hybrid modality that generates images by illuminating tissue with pulsed laser light. When tissue absorbs the light energy, it heats up and expands rapidly, producing broadband ultrasound waves - the photoacoustic effect. These sound waves are detected by a standard ultrasound transducer. The resulting image reflects the optical absorption properties of tissue rather than its acoustic echogenicity, providing fundamentally different tissue contrast that may distinguish cancer from benign tissue more effectively.
Photoacoustic imaging combines the best of both optical and acoustic modalities: it achieves the tissue contrast advantages of optical imaging while retaining the depth penetration and spatial resolution of ultrasound. It can image optical absorption properties at depths of several centimeters with submillimeter resolution - capabilities that pure optical imaging cannot achieve due to rapid loss of resolution with depth in biological tissue.
Different tissue constituents - including oxyhemoglobin, deoxyhemoglobin, lipid, fat, melanin, and collagen - have distinct light absorption spectra in the near-infrared region (700-1000 nm). Multiwavelength photoacoustic imaging acquires images at multiple laser wavelengths, each tuned to the absorption peak of a specific tissue chromophore. By comparing the photoacoustic signal intensity across wavelengths, researchers can characterize the biochemical composition of tissue without physical contact or biopsy.
This study focused on two wavelengths chosen to target hemoglobin - the dominant chromophore in the 700-1000 nm range. Laser wavelengths of 760 nm (where deoxyhemoglobin has peak absorption) and 800 nm (the isosbestic point where deoxyhemoglobin and oxyhemoglobin have equal absorption) were analyzed. Cancer is associated with excess blood vessel formation - angiogenesis - meaning malignant tissue contains more hemoglobin than normal tissue, producing stronger photoacoustic signals and different spectral characteristics.
Instead of simply comparing photoacoustic signal amplitudes (grayscale pixel values), this study applied frequency domain analysis to extract more information from the photoacoustic signal. The radiofrequency signal generated at each imaging point was analyzed to compute its power spectrum - the distribution of signal energy across frequencies. Three spectral parameters were extracted: slope (how the power spectrum changes with frequency), midband fit (the power level at the center frequency), and intercept (the power level extrapolated to zero frequency). These parameters encode information about the physical dimensions of the photoacoustic absorbers in addition to their optical absorption properties.
The physical basis for this approach comes from the relationship between absorber size and spectral characteristics: mathematically, slope depends only on absorber geometry (not absorption coefficient), while midband fit and intercept depend on both absorber size and optical absorption. Frequency domain analysis thus provides complementary information - slope reflects microstructural dimensions of the absorbers, while midband fit and intercept reflect both structure and biochemical composition, potentially capturing more aspects of the tissue differences between malignant and normal prostate.
This was an ex vivo experimental study using freshly excised human prostate specimens from 30 patients who underwent radical prostatectomy for biopsy-confirmed prostate cancer. Immediately after surgery, a genitourinary pathologist sectioned each excised prostate into thin slices (2-5 mm thick, 2-4 cm wide) and selected a section with a grossly visible nodule for photoacoustic imaging. The entire process from surgery to imaging was completed within one hour to preserve tissue freshness.
The photoacoustic imaging system used transillumination geometry: the pulsed laser (wavelength tunable from 700-1000 nm, 10 Hz pulse rate, 5-nanosecond pulse duration) illuminated the prostate specimen from one side while a 32-element ultrasound transducer array on the opposite side detected the resulting photoacoustic waves. An acoustic lens focused the photoacoustic waves onto the transducer array. Raster scanning with dual-axis linear motors allowed three-dimensional data acquisition. Laser intensity was maintained at approximately 5 mJ/cm2, well below safe human exposure limits.
The histopathologic slide from each specimen, analyzed by the genitourinary pathologist and used as the ground truth, was scanned and manually co-registered with the photoacoustic C-scan image (a cross-sectional image at a specific depth plane) to identify regions of interest (ROIs) corresponding to malignant cancer, benign prostatic hyperplasia (BPH), and normal prostate tissue. A total of 53 ROIs were analyzed across the 30 patients: 19 malignant, 8 BPH, and 26 normal.
For each ROI, the raw radiofrequency photoacoustic signals were corrected for wavelength-dependent laser intensity variation and depth-dependent light attenuation using the Beer-Lambert law. Signals were then windowed using Hamming windows (1 microsecond per window, 30% overlap) and transformed into power spectra using fast Fourier transform. The spectra were calibrated by dividing out the transducer transfer function to remove equipment artifacts, then fitted with a straight line to extract slope, midband fit, and intercept.
Statistical analysis using two-sample, two-tailed t-tests showed that all four parameters (slope, midband fit, intercept, and photoacoustic pixel value) were significantly different (P less than 0.05) between malignant and normal prostate tissue at both wavelengths - giving 8 out of 8 significant comparisons. For the clinically relevant malignant versus nonmalignant grouping (cancer versus all non-cancer tissue combined), all 8 parameters across both wavelengths were again significantly different, with p-values reaching as low as 5.53 x 10^-56.
The malignant versus BPH comparison was the most challenging: only 5 of 8 parameters were significantly different. This reflects the biological similarity between these two tissue types in terms of increased cellular density and vascular changes - BPH involves increased gland size and blood supply, partially mimicking some features of cancer. However, the parameters that did differ significantly between these categories still suggest photoacoustic spectral analysis can provide meaningful discrimination even in this difficult comparison.
Among the four parameters, midband fit showed the best overall performance: it was significantly different for all four tissue category pairs (malignant vs. normal, malignant vs. BPH, malignant vs. nonmalignant, BPH vs. normal) at both wavelengths, with consistently very small p-values. This makes midband fit the single most informative parameter for tissue discrimination in this dataset, combining sensitivity to both optical absorption and absorber size.
Comparing the two wavelengths, 800 nm outperformed 760 nm: at 800 nm, all three spectral parameters differentiated BPH from normal tissue significantly, while at 760 nm, slope failed to achieve significance for that comparison. For the key malignant versus normal and malignant versus nonmalignant comparisons, both wavelengths performed equally well. The overall superiority of 800 nm relates to the isosbestic point property, where equal contributions from deoxy- and oxyhemoglobin may provide more stable tissue characterization.
The higher photoacoustic signal amplitude (pixel values) in malignant tissue reflects tumor angiogenesis: cancer cells stimulate the formation of new blood vessels to supply their rapid growth, increasing the hemoglobin content of the tissue. Since hemoglobin is a primary photoacoustic absorber in the near-infrared range, malignant tissue produces stronger photoacoustic signals than normal tissue with less vascularity. The significantly higher midband fit and intercept values in malignant tissue confirm this absorption-driven effect.
The spectral parameter slope behaves differently from the others: mathematically, slope depends only on the geometry of the photoacoustic absorbers (their size and distribution), not on optical absorption. The lower slope values in malignant compared to normal tissue suggest that the characteristic microstructural scale of the absorbing elements is different in cancer - possibly reflecting differences in microvessel caliber, vessel density patterns, or cell size that alter the dominant absorber dimensions.
Midband fit and intercept depend on both optical absorption and absorber geometry, making them more informationally rich than either pixel values (absorption only) or slope (geometry only). This combined dependence is likely why midband fit emerged as the best single discriminating parameter - it captures the full multidimensional difference between tissue types rather than just one aspect of the photoacoustic signal.
This ex vivo study establishes the technical feasibility of photoacoustic spectral analysis for prostate tissue differentiation, but several challenges must be overcome before in vivo clinical use. In a living patient, photoacoustic imaging would need to switch from transillumination geometry (laser on one side, detector on the other) to reflection mode (laser and detector on the same side), since the prostate cannot be accessed from both sides during a transrectal procedure. This makes noninvasive laser delivery to the prostate challenging.
The light attenuation correction model used in this ex vivo study was a simplified one-dimensional Beer-Lambert correction appropriate for thin tissue slices. For in vivo imaging, overlying tissue between the light source and the prostate would absorb and scatter laser light in complex three-dimensional patterns, requiring more sophisticated optical inversion algorithms to accurately quantify local light fluence and recover accurate spectral parameters. The presence of overlying tissue may also shift the absolute values of spectral parameters compared to ex vivo measurements.
Despite these limitations, the frequency range analyzed (2.4-7.4 MHz) is fully compatible with clinical ultrasound imaging frequencies, suggesting that photoacoustic systems could be integrated into existing transrectal ultrasound platforms. This integration path - where photoacoustic capability is added to the ultrasound system already in use during biopsy - represents a practical clinical translation route that leverages existing diagnostic infrastructure and would not require entirely new equipment.
This study demonstrates that frequency domain analysis of multiwavelength photoacoustic signals can successfully differentiate malignant prostate cancer from both benign prostatic hyperplasia and normal prostate tissue in freshly excised human specimens. All three spectral parameters and photoacoustic pixel values showed statistically significant differences between malignant and nonmalignant prostate tissue at both 760 and 800 nm wavelengths.
The finding that midband fit was the most consistently discriminating parameter across all tissue category pairs - combining sensitivity to both optical absorption and absorber microstructure - provides specific guidance for future signal processing optimization. The superior performance at 800 nm compared to 760 nm identifies the preferred wavelength for future system development.
Key limitations include the small sample size of 30 patients and 53 ROIs, the ex vivo experimental setting that does not fully replicate in vivo tissue physiology, and the absence of classification accuracy metrics such as sensitivity, specificity, and AUC. Future work should expand the patient cohort, address the technical challenges of in vivo photoacoustic probe design, and test whether spectral analysis can guide real-time biopsy targeting to improve prostate cancer detection rates over standard systematic TRUS-guided biopsy.