Frequency Domain Analysis of Multiwavelength Photoacoustic Signals for Differentiating Among Malignant, Benign, and Normal Thyroids in an Ex Vivo Study With Human Thyroids

Abstract

Objectives: This study investigated the capability of spectral parameters, extracted by frequency domain analysis of photoacoustic signals, to differentiate among malignant, benign, and normal thyroid tissue.

Methods: We acquired multiwavelength photoacoustic images of freshly excised thyroid specimens collected from 50 patients who underwent thyroidectomy after having a diagnosis of suspected thyroid lesions. A thyroid cytopathologist marked histologic slides of each tissue specimen. These marked slides were used as ground truth to identify the regions of interest (ROIs) corresponding to malignant, benign, and normal thyroid tissue. Three spectral parameters: namely, slope, midband fit, and intercept, were extracted from photoacoustic signals corresponding to different ROIs.

Results: Spectral parameters were extracted from a total of total of 65 ROIs. According to the ground truth, 12 of 65 ROIs belonged to malignant thyroids; 28 of 65 ROIs belonged to benign thyroids; and 25 of 65 ROIs belonged to normal thyroids. Besides slope, the other 2 spectral parameters and grayscale photoacoustic image pixel values were found to be significantly different (P < .05) between malignant and normal thyroids. Between benign and normal thyroids, all 3 spectral parameters and photoacoustic pixel values were significantly different (P < .05).

Conclusions: Preliminary results of our ex vivo human thyroid study show that the spectral parameters extracted from radiofrequency photoacoustic signals as well as the pixel values of 2-dimensional photoacoustic images can be used for differentiating among malignant, benign, and normal thyroid tissue.

Keywords: frequency domain analysis; head and neck ultrasound; photoacoustic imaging; thyroid cancer; tissue characterization.

Figure 1

Ex-vivo PA imaging system

Figure 2

Ex-vivo C-scan PA images of a freshly excised thyroid tissue specimen acquired at three different wavelengths. (A) Thyroid tissue specimen (B) Histology slide of the thyroid specimen (C) Histology slide overlaid on the thyroid tissue specimen (D) PA image acquired at 760 nm (E) PA image acquired at 800 nm (F) PA image acquired at 850 nm. White circles in A, D, E, F and blue circles in B, C show ROI corresponding to malignant tissue.

Figure 3

Power spectrum analysis on a single A line PA signal (A) C scan PA image of the thyroid specimen shown in Figure 2A (B) PA A line signal generated by the tissue corresponding to the blue colored pixel (C) PA signal in Figure 3B is windowed [multiplied] by a Hamming window of the same length (D) Normalized amplitude spectrum of PA A-line signal in Figure 3C and one way transducer spectrum in the usable bandwidth region [2.4 MHz – 7.4 MHz] (F) Calibrated power spectrum (black line) along with the best fit straight line (red line) fitted to the calibrated power spectrum.

Figure 3

Power spectrum analysis on a single A line PA signal (A) C scan PA image of the thyroid specimen shown in Figure 2A (B) PA A line signal generated by the tissue corresponding to the blue colored pixel (C) PA signal in Figure 3B is windowed [multiplied] by a Hamming window of the same length (D) Normalized amplitude spectrum of PA A-line signal in Figure 3C and one way transducer spectrum in the usable bandwidth region [2.4 MHz – 7.4 MHz] (F) Calibrated power spectrum (black line) along with the best fit straight line (red line) fitted to the calibrated power spectrum.

Figure 3

Power spectrum analysis on a single A line PA signal (A) C scan PA image of the thyroid specimen shown in Figure 2A (B) PA A line signal generated by the tissue corresponding to the blue colored pixel (C) PA signal in Figure 3B is windowed [multiplied] by a Hamming window of the same length (D) Normalized amplitude spectrum of PA A-line signal in Figure 3C and one way transducer spectrum in the usable bandwidth region [2.4 MHz – 7.4 MHz] (F) Calibrated power spectrum (black line) along with the best fit straight line (red line) fitted to the calibrated power spectrum.

Figure 3

Power spectrum analysis on a single A line PA signal (A) C scan PA image of the thyroid specimen shown in Figure 2A (B) PA A line signal generated by the tissue corresponding to the blue colored pixel (C) PA signal in Figure 3B is windowed [multiplied] by a Hamming window of the same length (D) Normalized amplitude spectrum of PA A-line signal in Figure 3C and one way transducer spectrum in the usable bandwidth region [2.4 MHz – 7.4 MHz] (F) Calibrated power spectrum (black line) along with the best fit straight line (red line) fitted to the calibrated power spectrum.

Figure 3

Power spectrum analysis on a single A line PA signal (A) C scan PA image of the thyroid specimen shown in Figure 2A (B) PA A line signal generated by the tissue corresponding to the blue colored pixel (C) PA signal in Figure 3B is windowed [multiplied] by a Hamming window of the same length (D) Normalized amplitude spectrum of PA A-line signal in Figure 3C and one way transducer spectrum in the usable bandwidth region [2.4 MHz – 7.4 MHz] (F) Calibrated power spectrum (black line) along with the best fit straight line (red line) fitted to the calibrated power spectrum.

Figure 3

Power spectrum analysis on a single A line PA signal (A) C scan PA image of the thyroid specimen shown in Figure 2A (B) PA A line signal generated by the tissue corresponding to the blue colored pixel (C) PA signal in Figure 3B is windowed [multiplied] by a Hamming window of the same length (D) Normalized amplitude spectrum of PA A-line signal in Figure 3C and one way transducer spectrum in the usable bandwidth region [2.4 MHz – 7.4 MHz] (F) Calibrated power spectrum (black line) along with the best fit straight line (red line) fitted to the calibrated power spectrum.

Figure 4

Average values of the power spectral parameters namely slope, midband fit, intercept along with pixel values corresponding to malignant, benign and normal thyroid tissue at three different wavelengths (A) Average slope values (B) Average midband fit values (C) Average intercept values and (D) Average PA pixel values. Error bars show the standard deviation of the corresponding parameter over the particular tissue types.

Figure 5

Average values of the power spectral parameters namely slope, midband fit, intercept along with pixel values corresponding to malignant and nonmalignant thyroid tissue at three different wavelengths (A) Average slope values (B) Average midband fit values (C) Average intercept values and (D) Average PA pixel values. Error bars show the standard deviation of the corresponding parameter over the particular tissue types.

Figure 5

Average values of the power spectral parameters namely slope, midband fit, intercept along with pixel values corresponding to malignant and nonmalignant thyroid tissue at three different wavelengths (A) Average slope values (B) Average midband fit values (C) Average intercept values and (D) Average PA pixel values. Error bars show the standard deviation of the corresponding parameter over the particular tissue types.

Figure 6

Comparison of the four parameters using two sample two tailed student t test between pairs of different thyroid tissue types (A) Malignant vs normal (B) Malignant vs benign (C) Benign vs normal (D) malignant vs nonmalignant. The red line in each plot show the level corresponding to the decision criterion (p = 0.05). A p-value below 0.05 indicates that the corresponding spectral parameter is likely to be highly efficient in differentiating between a given pair of tissue type. For example in order to differentiate between malignant and normal tissue in Figure 6A, three parameters, midband fit, intercept and pixel value at all three wavelengths are deemed highly efficient but slope is not.