In biomedical onstudies, the characterization of molecular and functional information about the underlying tissue can significantly improve theour intuition in analyzing the morphology ones d, treatment efficacy, and metabolites of target tissues. Among many biomedical imaging techniques, optical imaging methods have been widely applied for small animal studies due to their cost-efficiency, ease of implementation, non-ionizing radiation, and real-time imaging capability [1]. More importantly, optical imaging techniques can also provide molecular and functional information by tuning the wavelength of the light source. While advantageous, the strong optical diffusion of the pure optical imaging techniques in biological tissues leads to a reduced penetration depth, thus limiting clinical translation.

Photoacoustic imaging (PAI), one of the branches of optical imaging, provides the added advantage of increased imaging depth [2]. Compared to other optical imaging techniques, PAI inherits ultrasound imaging characteristics (USI), which increases its ability to visualize structural information in deep tissue. The signal generation in PAI is based on the photoacoustic (PA) effect, which is energy transduction from light to ultrasound (US) [3]. In brief, PA images can be achieved through the following procedure: (i) pulse laser illumination, (ii) light absorption by chromophores, (iii) momentary heat generation, (iv) acoustic wave (for i.example., PA wave) generation through thermoelastic expansion, (v) signal detection by US transducer, and (vi) image generation. The resulting PA images, formed from the acoustic wave, include the optical absorption characteristics of the underlying biological tissue. Thus, PA images can provide molecular and functional information with a good ultrasound resolution in deep tissue ^[4][5][4,5]. In addition to the endogenous chromophores, contrast-enhanced PAI ^[6][7][6,7] has been focstusdied by developing exogenous contrast agents including organic dyes ^{[8][9][10][11][12]}[8,9,10,11,12] and inorganic nanoparticles ^{[13][14][15][16][17][18]}[13,14,15,16,17,18]. Recently, contrast agents that absorb light in the second near-infrared region (NIR-II, 1000–1350 nm) have been investigated. In NIR-II, a greater imaging depth can be achieved with reduced tissue scattering and background noise compared to the first near-infrared region (NIR-I, 650–950 nm), which is mainly used for contrast-enhanced PAI ^{[19][20][21][22][23][24][25]}[19,20,21,22,23,24,25].

One unique advantage of PAI is its scalable resolution and large imaging depth for the target region [26]. Since laser excitation can be tightly focused in shallow areas, high-resolution PA images can be achieved within the optical diffusion limit (~1 mm under the skin) ^[27][28][27,28]. High-resolution PAI has been used for visualizing the hemodynamics of the brain, ear, and eye of mice in vivo ^[29][30][29,30]. Since the light is diffused beyond the optical diffusion limit for deeper imaging depths, the US transducers determine the resolution at a greater depth ^{[31][32][33][34]}[31,32,33,34].

For clinical onresearch, PAI platforms have been developed and applied. Among them, the Vintage series (Verasonics, Kirkland, WA, USA), which equips the most-advanced programmable platform for designing user-defined operation sequence, is widely used ^[35][36][35,36]. The VevoLAZR series (FujiFilm VisualSonics, Toronto, ON, Canada) is another widespread commercial system for PAI research ^[37][38][37,38]. The main advantage of this system is a user-friendly interface, with real-time imaging and spectral analysis capability. Its high-frequency US transducer can provide high-resolution images, but it also limits the application area to a shallow area which is not favorable for clinical translation. The MSOT Acuity series (iThera Medical, Munich, Germany) has also been applied in clinical trials with multispectral PA analyses ^[39][40][39,40]. Its arc-shaped array can provide volumetric images, but its relatively small field of view is not suitable for general clinical applications that require a large imaging area. An FDA-cleared US machine (EC-12R, Alpinion Medical Systems, Anyang, Korea) has also been used to develop a clinical PAI system ^[41][42][41,42]. From the programmable platform in the US machine, the user can design their own operation sequence for their specific application.

Various geometry of multi-element array transducers has been developed and used for the clinical translation of PAI ^{[43][44][45][46][47]}[43,44,45,46,47]. In typical clinical PAI systems, both US and PA images are acquired by controlling the data acquisition sequence [48]. The dual-modal images complement each other by visualizing molecular and functional information in the PAI and specific morphologies in the USI [49]. Dual-modal PA and US imaging (PAUSI) has been used for clinical investigation in humans ^[50][51][50,51]. The multispectral PA responses provide metabolomic information about the biological tissues, thus indirectly providing valuable information about the cancerous tissues ^[52][53][52,53].

2. Multispectral Photoacoustic Analysis of Thyroid Gland

Thyroid cancer is one of the most common cancers, with an increasing global incidence rate in men and women ^[54][55][56][63,64,65]. The gold standard for diagnosing thyroid nodules is fine-needle aspiration biopsy (FNAB) ^[57][66]. The triaging for FNAB of the nodule is determined by the characteristics of nodules in USI ^[58][59][60][67,68,69]. Although the sensitivity of US-guided triaging is greater than 90%, the lack of functional metabolomics results in a low specificity of 20–50% ^[61][70]. The high false-positive rate leads to unnecessary FNAB, which results in the over-diagnosing of the tumor. Thus, clinical trials have been conducted to enhance the accuracy of triaging thyroid nodules using PAI due to its molecular and functional imaging capability. Dogra et al. analyzed 88 resected tissues (13 malignant nodules, 30 benign nodules, 13 colloid accumulations, and 32 normal tissues) from 50 patients (11 malignant and 39 benign) ^[62][71]. Four different wavelengths (760, 850, 930, and 970 nm) were used for the spectral unmixing of HbO, HbR, lipid, and water components from multispectral PA data. Statistically significant differences were found in HbO and HbR between malignant and other types of tissues. In particular, HbR components were significantly different between malignant and normal tissue, with a p-value of 0.003 in the student t-test. The results showed the promising feasibility of PA-guided classification with a sensitivity of 69.2% and a specificity of 96.9%, but this study was limited to ex vivo environments only. Thus, for clinical translation, further in vivo validation is needed. Dima et al. demonstrated the in vivo imaging capability of their PA and US system for the human thyroid ^[63][72]. They recruited two healthy volunteers to acquire PA images with a single excitation wavelength of 800 nm. US Doppler images were also acquired in the same region to verify the blood vessel’s position. By comparing the PA images with the US Doppler images, surrounding blood vessels extending from the isthmus and carotid artery to the anterior of the thyroid gland were identified. The results showed the feasibility of in vivo PAI using the arc array US transducer by confirming the matched positions of blood vessels (white arrows in. However, the spectral analysis of cancerous nodules was not available in this study. Yang et al. compared in vivo PA images between papillary thyroid cancer (PTC) patients and healthy volunteers ^[64][73]. Although they achieved PA responses from cancerous nodules, the number of patients (10 PTC and 3 normal) included in the study was insufficient for statistical analysis. In addition, multispectral analysis was also not available in this study because they used a single excitation wavelength of 1064 nm. Roll et al. presented multispectral PA analyses for differentiating tissue disorders in the thyroid gland ^[65][74]. The composition of HbO, HbR, fat, and water were spectrally unmixed from the in vivo PA images of the enrolled patients (6 Graves’ disease, 3 malignant, 13 benign, and 8 healthy), obtained using eight excitation wavelengths (700, 730, 760, 800, 850, 900, 920, and 950 nm). The sO₂ levels of the thyroid were also visualized and investigated. The contours of the thyroid glands were determined by the corresponding US images. Statistical analyses demonstrated significant differences between diseased and normal thyroid tissues. Recently, Kim et al. presented a multispectral PA analysis with a statistically sufficient number of samples (23 PTC and 29 benign), the largest number of patients in a clinical thyroid onestudy to date ^[66][75]. They achieved multispectral PAI using five wavelengths (700, 756, 796, 866, and 900 nm). The corresponding US data were also acquired simultaneously for delineating the boundary of nodules. Similar to the previous onstudies, the sO₂ levels in nodules were acquired through the spectral unmixing of HbO and HbR. Three parameters were quantified and used for training the decision function: (i) PA spectral gradient: the slope of a first-order polynomial fitted line to the average value of the top 50% of PA signals within the nodule boundary at each wavelength; (ii) relative sO₂ level: the average value of the top 50% sO₂ values within the nodule; (iii) skewed angle of sO₂ distribution: the skewed angle of the Gaussian-fitted distribution for the top 50% of sO₂ values within the nodule. With the values of the three parameters scattered in a 3D plane, a support vector machine was trained to determine the 3D decision boundary, which showed a good classification accuracy with a sensitivity of 78% and a specificity of 93%. The classification accuracy was further enhanced using a novel scoring method (ATAP score), which combined a conventional USI-based scoring method (for i.example., ATA guideline score) and the photoacoustic probability of malignancy. The novel scoring method improved the sensitivity to 83% and the specificity to 93%. Thus, the results showed a great potential for enhancing the triaging accuracy of thyroid nodules using a multiparametric analysis of multispectral PA data as a complementary method to the conventional triaging method. While PA analyses of thyroid nodules have been conducted by various groups worldwide, the validation of multispectral PA analysis is still at the initial stage of evaluation. Further studit wases are required to address the following issues for successful clinical translation. (i) Larger number of patients are needed to enhance the reliability of this technique. (ii) In addition to PTC, the classification of other types of thyroid cancers such as follicular, medullary, and anaplastic thyroid cancers would expand the application area. (iii) Quantitative analyses of PA responses in skin color are needed. (iv) System improvement with a deeper imaging depth, faster frame rate, and smaller size would enhance the image quality for multispectral analyses.