Enhancing the pathological diagnosis of oral cancers using broadband ultraviolet photoacoustic microscopy

Abstract

Conventional histopathological diagnostic methods, such as hematoxylin-eosin (H&E) staining and immunohistochemistry (IHC), face significant challenges in intraoperatively diagnosing early-stage oral cancer and certain cancer types. To address this, we propose a novel pathological diagnostic method based on a broadband ultraviolet photoacoustic microscope (bUV-PAM). This method capitalizes on the unique optical absorption properties of nuclei and high-frequency photoacoustic spectral variations, enabling the simultaneous performance of photoacoustic morphological imaging (PAMI) and photoacoustic spectral analysis (PASA) within 20 minutes. As a result, photoacoustic images of normal and cancerous mouse tissues, including oral squamous cell carcinoma (OSCC) and adenoid cystic carcinoma (ACC), demonstrate strong consistency with their corresponding H&E images. Moreover, the photoacoustic spectral profiles of these tissues exhibit significant differences, validating the effectiveness of PASA in differentiating between normal and cancerous states. Furthermore, PASA rapidly identifies diseases that are challenging to diagnose with H&E staining, such as the transition from normal mouse mucosa tissue to OSCC and the differentiation between OSCC and ACC, as verified by the more time-consuming IHC method. Consequently, the photoacoustic pathological diagnosis method enhances current diagnostic practices, promising to offer crucial support for the rapid intraoperative identification of human early-stage oral cancers and various cancer types.

1. Introduction

Oral cancer is the sixteenth-most prevalent malignant tumor worldwide, with increasing morbidity and mortality rates [1]. Oral Squamous Cell Carcinoma (OSCC), which constitutes 90%–95% of oral cancers [2], is the most frequent type, while Adenoid Cystic Carcinoma (ACC) is another common form [3]. Delayed diagnosis significantly contributes to oral cancer mortality because early-stage nonmetastatic tumors are more amenable to surgical resection compared to advanced cancers with extensive metastases and poor prognoses [4]. Accurate differentiation between OSCC and ACC is crucial due to the distinct treatment strategies required, underscoring the importance of precise diagnostic approaches [5]. The management of OSCC primarily involves surgery, radiotherapy, and chemotherapy. In contrast, treatment for ACC necessitates a more complex regimen that includes targeted therapies and immunotherapy [6]. Consequently, early cancer identification and differentiation of various cancer subtypes are essential for improving cure rates and patient survival.

Advancements in biomedical imaging and detection technologies have significantly enhanced the diagnosis of oral cancer. Intraoral camera systems [7,8] allow non-invasive lesion visualization on the oral mucosa surface but are limited in detecting deeper tissue abnormalities. Depth imaging procedures, such as oral Magnetic Resonance Imaging (MRI) [9,10] and oral Computed Tomography (CT) [11,12], excel at identifying lesions within deeper tissues but struggle with resolution constraints in detecting smaller lesions. Moreover, these imaging methods generally offer a general assessment of lesions and fail to deliver an accurate pathological diagnosis. Molecular biology approaches [13,14], which analyze changes in biomarkers, genetic mutations, and protein expression levels in saliva, blood, or exfoliated oral mucosal cells, show promise in accurately detecting lesions. However, these techniques involve time-consuming sample collection and testing processes and cannot pinpoint the lesion's location, limiting clinical intraoperative guidance.

Hematoxylin and Eosin (H&E) staining, although considered the gold standard for oral cancer pathological diagnosis, has certain limitations [15]. First, H&E staining primarily reveals nuclear morphology, distribution, and structure, which are typically evident only in the advanced stages of tumors, posing challenges for early cancer diagnosis. Second, due to the similarities in the affected locations, metastatic patterns, and clinical symptoms of OSCC and ACC, H&E staining images may confuse distinguishing between these two cancer types. Moreover, ACC may occasionally masquerade as OSCC or form a collision tumor [16,17], further clouding the diagnostic process. To overcome these challenges, pathologists routinely employ Immunohistochemical (IHC) to identify specific antigens within tumor cells, thereby facilitating early diagnosis and accurate differentiation among various cancer types. For instance, the Ki67 antibody is commonly used to assess proliferative activity in OSCC [18,19]; P63 and Cytokeratin-5 (CK5) antibodies aid in diagnosing OSCC [20,21]; Cytokeratin-7 (CK7) and Cytokeratin-8 (CK8) antibodies help identify ACC [22,23]. Despite the efficacy of IHC, the preparation typically extends over several days, making it impractical for intraoperative diagnosis. Moreover, IHC often requires the simultaneous labeling of multiple antigen proteins, necessitating more clinical samples and higher costs. Consequently, there is an urgent need to develop a rapid and precise pathological diagnostic method to enhance the diagnosis of oral tumors.

Ultraviolet Photoacoustic Microscopy (UV-PAM) offers significant promise for rapid, label-free histopathological diagnosis [24–26]. The high optical absorption of the abundant DNA/RNA in the nuclei at a 266 nm ultraviolet laser allows UV-PAM to produce photoacoustic images that closely resemble H&E staining images without additional labels. Optical resolution photoacoustic microscopy (OR-PAM) utilizes focused ultraviolet lasers to excite photoacoustic waves, achieving cellular or subcellular lateral resolution. Various tissues, including the human breast [24], colorectal [25], and bone [26], have been imaged label-free using photoacoustic techniques, yielding images resembling H&E staining images. Given the high resemblance between photoacoustic and H&E staining images, photoacoustic technologies have the potential to diagnose most oral tumors. Nonetheless, photoacoustic imaging may encounter challenges in early-stage tumor screening and differentiating certain cancer types.

Fortunately, in addition to providing diagnostic features similar to H&E staining, UV-PAM offers enhanced diagnostic insights via the photoacoustic spectra derived from the Fourier transformation of time-domain photoacoustic signals. These spectral characteristics are associated with nuclei's geometric parameters and physical properties, including size, shape, acoustic velocity, and density. Thus, precise quantification of the photoacoustic spectra can significantly distinguish biological particles with varying properties [27,28]. For instance, broadband optical microresonators can identify particles with diverse compositions, sizes, and structures, thus aiding in identifying various microbial cells [27]. A broadband surface plasma resonance sensor can effectively distinguish between particles with similar shapes but differing acoustic velocities and densities [28]. These studies highlight how the photoacoustic spectrum can quantify geometric characteristics, such as the size and shape of biological particles. Furthermore, it can assess physical properties, such as acoustic velocity and density of biological entities, thereby offering detailed pathological information that is typically elusive with traditional morphological imaging techniques like H&E staining and standard biomedical optical imaging, and underscoring its potential to enhance pathological diagnostics.

In this study, we utilized a broadband UV-PAM (bUV-PAM) system based on optical evanescent wave sensing to diagnose various pathological conditions in mouse oral tissues within 20 min, generating dual-mode virtual slices simultaneously involving Photoacoustic Morphological Imaging (PAMI) and Photoacoustic Spectral Analysis (PASA). The PAMI results exhibited high consistency with the H&E staining images, enabling label-free differentiation between normal tongue mucosa tissues, normal parotid gland tissues, and their malignant counterparts. In addition, we employed PASA to achieve a significant distinction between the aforementioned normal tissue and the corresponding cancerous tissue, preliminarily validating the feasibility of photoacoustic spectral diagnosis of oral cancer. For oral tumor diseases that pose diagnostic challenges with H&E staining or PAMI, we employed PASA for further diagnosis and IHC for verification. As a result, PASA accurately identified six progressive stages, from normal oral mucosa tissue to early-stage cancer to advanced OSCC, underscoring the importance of early intraoperative diagnosis of oral cancer. Furthermore, PASA effectively differentiated OSCC and ACC tissues, suggesting the potential to distinguish between different types of oral cancers. Consequently, the novel photoacoustic pathological method emerged as a promising tool for enhancing current oral cancer pathological diagnosis, providing vital support for the rapid and precise identification of various oral cancers.

2. Results

2.1. bUV-PAM system integrating an optical evanescent wave sensor with an ultraviolet illumination laser

Figure 1 shows our previously developed bUV-PAM system [29], optimized for diagnosing oral lesions. Due to the high optical absorption coefficient of DNA/RNA under ultraviolet illumination, bUV-PAM can precisely extract photoacoustic signals from nuclei without additional labeling. For photoacoustic excitation, an ultraviolet pulsed laser (Bright Solutions, Wedge HF 266, wavelength 266 nm, 1–50 kHz repetition frequency) was focused on the sample via an ultraviolet objective lens (Newport, U-13X-LC, NA = 0.13), generating a diffraction-limited lateral resolution of approximately 1.25 µm. In our previous work, we established that the lateral resolution of the bUV-PAM is approximately 1.68 µm, as demonstrated by imaging a sharp blade edge. Cell nuclei within oral tissues, typically 5–10 µm in size [30,31], produce photoacoustic signals with bandwidths extending to hundreds of MHz [32,33]. The generation of photoacoustic signals from the nuclei is detailed in Section 3.1 of the following Ref. [29]. Hence, we employed a previously developed broadband optical evanescent wave sensor with a -6 dB bandwidth of ∼174 MHz [34] to capture these high-frequency photoacoustic waves. Specifically, a probing laser at 633 nm passes through optical components, including a polarizer (P₁), a half-wave plate (HWP), and a quarter-wave plate (QWP), to generate adjustable, elliptically polarized light. After passing through a quartz prism at a critical angle, the beam creates an optical evanescent wave on the lower surface of the prism to detect photoacoustic waves. The reflected detection light was divided into two identical beams by a beam splitter (BS) and then processed through orthogonal polarizers (P₂ and P₃), resulting in a time-varying linear polarized laser. Following a balanced photodetector (BPD), the output photoacoustic signal was amplified via a low-noise amplifier (ZHL-1-2W-S+, Mini-Circuits, 5–500 MHz) and recorded via a digital acquisition card (ATS9870, Alazar Technologies, 450 MHz bandwidth, 1 GS/s sampling rate). A label-free tissue section on a two-dimensional scanning platform (L401-01, Shanghai JY Tech Co., Ltd.) allowed morphological imaging of cell nuclei. Furthermore, real-time acquisition and analysis of the photoacoustic spectra provided comprehensive evaluations of tissue pathology by assessing the physical properties of cell nuclei.

Fig. 1.

Schematic representation of the bUV-PAM system integrating ultraviolet photoacoustic excitation and broadband optical evanescent wave sensing, enabling simultaneous obtaining of photoacoustic morphological images and performing photoacoustic spectral analysis. L: Lens; M: Mirror; OL: Objective lens; P: Polarizer; HWP: Half-wave plate; QWP: Quarter-wave plate; BS: Beam splitter; BPD: Balanced photodiode detector; X–Y stage: Two-dimensional scanning platform; PC: Personal computer; UV pulsed laser: Ultraviolet pulsed laser.

2.2. PAMI of normal and cancerous tissue sections

To validate the imaging capabilities of the bUV-PAM system on oral tissue sample sections, we performed PAMI on both normal and malignant tongue mucosa and parotid gland tissues. We then compared these photoacoustic images with the corresponding H&E staining images. Sequential sections were used for photoacoustic imaging and H&E staining to ensure high tissue consistency, as detailed in Section 3.1. Details on photoacoustic scanning, image reconstruction, and virtual staining are provided in Section 3.2.

Figures 2(a) and 2(b) present photoacoustic and corresponding H&E staining images of normal tongue tissue. Figures 2(c) and 2(d) offer detailed enlargements of the solid and dot-dash blue boxes in Figs. 2(a) and 2(b), respectively. A thorough comparison of the overview and magnified images revealed a striking similarity between the photoacoustic and H&E staining images. Figures 2(c) and 2(d) highlight the regular cellular distribution and the optimal nuclear-to-cytoplasmic ratio. The epithelial layer above the irregular blue dashed line includes prominent filiform papillae (arrow 1). Below this line, the lamina propria is marked by organized muscle fibers of the tongue (arrow 2), commonly seen in normal tissues [35]. Figures 2(e) and 2(f) display the photoacoustic and corresponding H&E staining images of OSCC, a malignant variant of normal tongue mucosa. A red dashed line demarcates the boundary between OSCC tissue (above) and normal tongue tissue (below). Enlarged views, shown in solid and dot-dash red boxes, are further detailed in Figs. 2(g) and 2(h). These Figures exhibit a strong consistency between the imaging techniques. Figures 2(g) and 2(h) illustrate the absence of filiform papillae (arrow 3) and disruptions in muscle fiber (arrow 4), which are indicative of malignant transformation (OSCC) [36]. To further validate the performance of PAMI, we quantified the nuclear-to-cytoplasmic ratio derived from H&E staining images and photoacoustic images of normal tongue mucosa tissue and OSCC, as illustrated in Fig. 2(i). The measured nuclear ratios were as follows: Normal-H&E 23.82 ± 5.452%, Normal-PA 21.47 ± 5.006%, OSCC-H&E 50.33 ± 7.135%, and OSCC-PA 50.05 ± 5.802%. These results demonstrate a high degree of consistency between H&E staining and photoacoustic images.

Fig. 2.

PAMI of normal and cancerous mouse tongue sections. (a)(b) Photoacoustic images and corresponding H&E staining images of normal tongue sections. (c)(d) Magnified regions indicated by the solid and dot-dash blue boxes in (a) and (b). Blue dashed lines separate the tongue mucosa epithelial layer (above) and the tongue mucosa lamina propria (below). Arrow 1 denotes filiform papillae, and arrow 2 points to regular tongue muscle fibers. (e)(f) Virtual staining photoacoustic images and corresponding H&E staining images of OSCC tissue sections. The red dashed line marks the boundary between normal tongue mucosa tissue (below) and OSCC tissue (above). (g)(h) Close-up views of the solid and dot-dash red boxes in (e) and (f). Arrow 3 indicates the absence of filiform papillae, while arrow 4 shows the disruption of the tongue muscle fibers. (i) The nuclear-to-cytoplasmic ratio observed in H&E staining images and photoacoustic images of normal tongue mucosa tissue and OSCC. SD: standard deviation.

Figures 3(a) and 3(b) present the photoacoustic and corresponding H&E staining images of the normal parotid tissue sections. Figures 3(c) and 3(d), the enlarged views of the blue dash and dot-dush boxes in Figs. 3(a) and 3(b), reveal an impressive consistency, illustrating regular cellular distribution and tissue architecture. Arrow 1 identifies the normal glandular lobule, and arrow 2 marks the salivary duct, which is characteristic of normal parotid tissues [35]. Conversely, Figs. 3(e) and 3(f) illustrate the photoacoustic and corresponding H&E staining images of ACC transformed from normal parotid tissue. Significant changes in Figs. 3 g and 3 h are evident in the enlarged views of the red dashed and dot-dash boxes in Figs. 3(e) and 3(f). These include the absence of normal glandular lobules and salivary ducts, a marked increase in nuclear density, the formation of cancer cell clusters (arrow 3), and localized necrosis at cluster centers (arrow 4), indicating progression to ACC [36]. The nuclear-to-cytoplasmic ratio obtained from H&E staining and photoacoustic imaging of normal salivary gland tissue and ACC are presented in Fig. 3(i). The measured nuclear ratios were as follows: Normal-H&E 16.53 ± 4.539%, Normal-PA 16.14 ± 4.957%, ACC-H&E 49.35 ± 9.314%, and ACC-PA 47.93 ± 10.78%. These results indicate a substantial agreement between H&E staining and photoacoustic images.

Fig. 3.

PAMI of normal parotid gland and ACC tissue sections in mice. (a)(b) Photoacoustic and corresponding H&E staining images of normal parotid gland tissue sections. (c)(d) Magnified views of the blue dash and dot-dush boxes in (a) and (b), respectively. Arrow 1 highlights the normal glandular lobule, while arrow 2 indicates the salivary duct. (e)(f) Photoacoustic and corresponding H&E staining images of ACC tissue sections. (g)(h) Enlarged views of the red dash and dot-dush boxes in (e) and (f), respectively. Arrow 3 points to cancer cell clusters, where normal acinar lobules and glandular ducts are absent. Arrow 4 points to localized central necrosis within the cancer cell clusters. (i) The nuclear-to-cytoplasmic ratio observed in H&E staining images and photoacoustic images of normal salivary gland tissue and ACC. SD: standard deviation.

These results highlight the potential of our bUV-PAM system in the label-free imaging of various oral samples, facilitating the diagnosis of oral pathology. Minor discrepancies between photoacoustic and H&E staining images may result from variations in nuclei distribution across adjacent tissue sections and potential damage and deformation during sample preparation.

2.3. PASA distinguishes between normal and cancerous tissue sections

To verify the potential of the bUV-PAM system to distinguish between normal and cancerous oral tissues, we first conducted PASA on normal mucosal and OSCC tissue sections. Figure 4(a) presents a schematic representation of the structural differences between normal oral mucosa and OSCC tissue. The left panel of Fig. 4(a) depicts the structured layers of normal oral mucosa, including the keratin (K), granular (G), spinous (S), and basal (B) layers, each exhibiting regular cell morphology and a well-defined arrangement. In contrast, the epithelial layer of OSCC is characterized by disordered cell arrangement, varied cell sizes and shapes, and an increased nuclear–cytoplasmic ratio. In Fig. 2, morphological photoacoustic images highlight the differences between normal oral mucosa and OSCC tissues. To confirm the efficacy of PASA for OSCC diagnosis, we performed PASA on both normal and OSCC tissue samples, as detailed in Section 3.3. Figure 4(b) illustrates the mean photoacoustic spectral variations for the 10 normal tongue and OSCC tissue groups. Normal tissue exhibits a -6 dB bandwidth of 127.9609 ± 0.4134 MHz, whereas OSCC displays a -6 dB bandwidth of 144.0781 ± 0.3071 MHz. Statistical analysis of these results, presented in Fig. 4(c), reveals a significant difference between the two stages, highlighting the bUV-PAM system's diagnostic potential.

Fig. 4.

PASA of normal mouse mucosa and OSCC tissue sections. (a) Schematic representation of the structural differences between normal oral mucosa and OSCC. E: Epithelium; B: Basal layer; S: Spinous layer; G: Granular layer; K: Keratin layer; BM: Basement membrane; LP: Lamina propria. (b) The average photoacoustic spectral distribution curve of normal oral mucosa and OSCC. (c) Statistical results of the -6 dB bandwidth of normal oral mucosa and OSCC tissue sections. SD: standard deviation.

Figure 5(a) shows a schematic depiction of the structural differences between normal parotid and ACC tissues. In normal parotid tissue, the ducto-acinar units are regularly arranged, exhibiting uniform cell morphology. In contrast, ACC is characterized by enlarged tumor cell nuclei with irregular shapes, and the ducto-acinar units appear disordered and obstructed. We examined 10 groups of normal parotid and ACC tissue using PASA, confirming its efficacy in identifying the cancerization of parotid tissue, as detailed in Section 3.3. Figures 5(b) and 5(c) demonstrate significant differences in the photoacoustic spectral characteristics between normal parotid and ACC tissue in mice. Notably, the -6 dB bandwidth increased from 131.8681 ± 0.5034 MHz in normal parotid tissue to 151.4042 ± 0.4022 MHz in ACC, indicating a clear distinction.

Fig. 5.

PASA of normal parotid and ACC tissue sections of mice. (a) Schematic representation of the structure of normal parotid gland and ACC. A: Acinus; I: Intercalated duct; S: Striated duct; E: Excretory duct; MEC: Myoepithelial cell. (b) The average photoacoustic spectral distribution curve of the normal parotid gland and ACC. (c) Statistical results for the -6 dB bandwidth of normal parotid gland and ACC tissue sections. SD: standard deviation.

The studies mentioned above and in Figs. 4 and 5 reveal significant photoacoustic spectral differences between pathological states, such as normal mucosal tissue versus OSCC and normal parotid tissue versus ACC. These differences may arise from substantial changes in nuclear physical properties, such as acoustic velocity, density, and elasticity, affecting the photoacoustic spectra [29]. By employing the broadband PASA method, we can effectively distinguish between normal and malignant oral tissues, enhancing the capability of label-free pathological diagnosis of oral cancer.

2.4. PASA distinguishes between different stages of OSCC

The progression from normal tissue to cancer, such as OSCC, occurs over an extended period. Typically, this progression follows several stages: edema injury stage (EG1), mild dysplasia (EG2), moderate dysplasia (EG3), carcinoma in situ and early invasive carcinoma (EG4), and invasive carcinoma (EG5) [37,38]. Rapid and precise identification of these lesions is essential for early tumor diagnosis and treatment. However, the pathological features of early-stage OSCC are not always apparent in H&E staining images [39,8]. Clinicians often rely on the IHC Ki67 biomarker to assess the developmental stages of OSCC by examining cellular proliferation [18,19].

Figure 6(a) shows the IHC Ki67 biomarker in the normal oral mucosa (CG), with brown granules restricted to the basal layer. Figure 6(b) depicts the IHC image of the edema injury stage (EG1), manifesting as damage in the keratinized layer, with Ki67 expression levels similar to those in normal mucosa (CG). Figure 6(c) illustrates the IHC image of the mild dysplasia (EG2), marked by initial positive Ki67 expression in basal and spinous layers, with increased brown granules in nuclei, signifying enhanced cell proliferation. Figure 6(d) shows the IHC image of the moderate dysplasia (EG3), where Ki67 expression extends to upper cellular layers, indicating abnormal growth. Figure 6(e) represents the IHC image of carcinoma in situ and early invasion (EG4), with Ki67 expression across all epithelial layers, reflecting uncontrolled proliferation. Lastly, Fig. 6(f) presents the IHC image of invasive cancer (EG5), with widespread Ki67 across epithelial layers, invading and disrupting normal tissue structure and function. While IHC can reveal the diagnostic characteristics of OSCC stages, the preparation process is time-consuming, posing challenges for intraoperative use.

Fig. 6.

PASA distinguishes between different stages of OSCC. (a)–(f) IHC staining for the Ki67 biomarker in oral mucosa sections with various progressive stages. (g) Average photoacoustic spectral distribution curves of the six stages. (h) Statistical results of photoacoustic spectral characteristics at -6 dB bandwidth across stages from normal buccal mucosa to OSCC. CG: Control group; EG: Experimental Group; EG1: Edema injury phase; EG2: Mild dysplasia; EG3: Moderate dysplasia; EG4: Carcinoma in situ and early invasive carcinoma; EG5: Invasive carcinoma. (▴ represents a comparison with other groups, and P < 0.001). SD: standard deviation.

During the progression of oral tumors, pathological changes can result in modifications to the physical properties of cell nuclei, such as density, acoustic velocity, and elasticity. Broadband PASA enables the detection of variations in these physical properties of biological particles, thereby providing valuable insights into tissue disease progression. To assess PASA's efficacy in diagnosing oral tumors at progressive stages, we employed the oral buccal mucosa to induce OSCC due to its accessibility and observable characteristics [40,41]. Using mouse buccal mucosa induced by chemical carcinogens, we modeled six progressive stages of OSCC. The preparations are detailed in Section 3.1, with success rates in Table 1. As the transitions from normal tissue (CG) to early-stage lesion grades (EG1–EG3), a notable contrast in photoacoustic spectral characteristics reflects the disease progression. Specifically, the -6 dB bandwidth increases from 128.4493 ± 0.5019 MHz in normal tissue (CG) to 129.9145 ± 0.8185, 130.4029 ± 0.4947, and 138.9499 ± 0.5052 MHz in early-stage lesions (EG1–EG3), respectively.

Table 1. Statistical characteristics of the mouse models.

Group	Number	Tumorigenesis
CG	10	0
EG1	10	5 (50%)
EG2	10	7 (70%)
EG3	10	7 (70%)
EG4	10	6 (60%)
EG5	10	6 (60%)

Upon progression to carcinoma in situ and early invasive carcinoma (EG4), the bandwidth rose significantly to 148.9621 ± 0.3442 MHz, marking cancer formation. This bandwidth further increased to 150.6716 ± 0.5453 MHz during the complete transition to the invasive carcinoma stage (EG5), indicating intensified cancer advancement. All differences are statistically significant (P < 0.05), as depicted in Figs. 6 g and 6 h. These results suggest that DNA replication from normal to cancerous states significantly alters the oral mucosa's physical characteristics. Broadband PASA effectively detects these changes, opening pathways for early-stage tumor detection, providing a foundation for diagnosing tumors at varying stages and guiding subsequent treatments.

2.5. PASA distinguishes between OSCC and ACC

In clinical practice, treatment strategies for oral cancer types differ significantly. For instance, OSCC is mainly treated with surgery, radiotherapy, and chemotherapy, whereas ACC necessitates more complex protocols, including targeted therapies and immunotherapy. Therefore, accurate differentiation of oral cancer types, such as OSCC and ACC, is essential for guiding clinical treatment strategies. While OSCC originates from squamous epithelial tissues and ACC originates from glandular epithelial tissues, their anatomical proximity can complicate distinction based on tumor location. In addition, the basaloid cells in OSCC exhibit notable morphological similarities to the tumor cells present in ACC, characterized by features including hyperchromasia nuclei and diminished cytoplasmic volume in H&E staining images. Moreover, both types share histological patterns such as nests, cords, adenoid, or cribriform structures [42]. Consequently, traditional H&E staining often faces challenges in differentiating specific oral cancer types, particularly in transitional areas such as the junctional regions of the oral mucosa and parotid gland. This can lead to uncertainties in clinical assessments, as H&E staining occasionally fails to clearly distinguish between OSCC and ACC.

To address these diagnostic challenges, pathologists use various IHC antibodies to precisely highlight antigens in tissue sections. For example, the combined use of P63 and CK5 aids in diagnosing OSCC, while CK7 and CK8 facilitate the diagnosis of ACC. The IHC biomarker results for OSCC and ACC are depicted in Figs. 7(a)–7(d). These Figures depict cellular heterogeneity and numerous multinucleated cells, affirming the tumorous nature of the samples. Specifically, Figs. 7(a) and 7(b) display positive P63 expression in the nucleus and CK5 expression in the cytoplasm, indicative of OSCC tissues. In contrast, Figs. 7(c) and 7(d) exhibit positive CK7 and CK8 staining primarily in the cytoplasm, which is characteristic of ACC tissues. However, the multiple antibodies staining process in IHC is time-consuming, posing challenges for intraoperative applications.

Fig. 7.

PASA distinguishes between OSCC and ACC. (a)–(d) IHC staining using different antibodies for diagnosing OSCC and ACC. (e) The average photoacoustic spectral distribution curve of OSCC and ACC. (f) Statistical comparison of -6 dB bandwidth in OSCC and ACC tissues. SD: standard deviation.

To validate the efficacy of the PASA method in accurately diagnosing oral cancer types, we focused on distinguishing between OSCC and ACC, as detailed in Section 3.3. Figures 7(e) and 7(f) reveal a -6 dB bandwidth of 144.0781 ± 0.3071 MHz for OSCC and 151.4042 ± 0.4022 MHz for ACC, highlighting a striking difference. These variations likely stem from the distinct nuclear properties of each cancer type effectively captured by PASA. Our study demonstrated PASA's potential in accurately discriminating oral cancers, complementing traditional H&E and IHC techniques in pathological diagnosis.

3. Materials and methods

3.1. Sample preparation

3.1.1. Establishment of mouse OSCC and ACC models

Healthy nude mice (4–6 weeks old) were maintained in specific pathogen-free (SPF) environments at the Institute of Radiation Medicine, Chinese Academy of Medical Sciences. To induce an OSCC, 2 × 10⁶ CAL27 cells per 100 µL phosphate-buffered solution (PBS, Solarbio, China) were injected into the tongue of a mouse. To establish an ACC model, 6 × 10⁶ SACC-LM cells per 200 µL PBS were injected into the left parotid gland. The control groups received equivalent volumes of PBS at the respective injection sites. Each experimental and control group consisted of 10 mice. Mice were monitored every three days for tumor progression and clinical symptoms, such as weight loss, difficulty eating, swelling, or ulcers at the injection sites. Mice were euthanized if their body weight decreased by 20% from the baseline. The diseased tissues were collected and fixed in 10% neutral buffered formalin (NBF) for subsequent processing, such as sectioning, staining, and IHC.

3.1.2. Establishment of progressive OSCC models

We employed a chemical carcinogen-induced method to generate samples representing progressive stages from normal tissue to OSCC. Each group comprised 10 male BALB/C nude mice (Vital River, 4–6 weeks, SPF). A carcinogen 7,12-dimethylbenz[a]anthracene (DMBA, MCE, China) solution in acetone, with a concentration of 5 g/L, was applied three times weekly to the left cheek pouch. The solution was dried entirely using an ear syringe until it was fully absorbed. This application procedure was repeated three times consecutively. Mice were humanely euthanized at 3, 6, 9, 12, and 15 weeks post-application to obtain five groups of mucosal samples representing different lesion stages. Tissue sections were diagnosed using IHC Ki67 biomarker images, and samples were characterized as follows: EG1, edema injury phase; EG2, mild dysplasia phase; EG3, moderate dysplasia phase; EG4, carcinoma in situ and early invasive carcinoma phase; and EG5, invasive carcinoma phase. The control group (CG), which received no treatment, was euthanized at each corresponding time point. All tissue samples were promptly fixed in 10% NBF for subsequent sectioning, staining, and IHC biomarkers.

3.1.3. Tissue sectioning, H&E staining, and the IHC biomarker

Fresh tissues were fixed in NBF for 48 hours, dehydrated in gradient alcohol, and embedded in paraffin to create paraffin blocks. Serial sections of 4-µm thickness were prepared to ensure uniform morphology and physical properties across different label-free sections. For photoacoustic acquisition, sections were mounted on quartz slides to eliminate any unwanted photoacoustic signals from the substrate under ultraviolet excitation. Given the high optical absorption of paraffin, it is essential to thoroughly remove paraffin from samples on quartz slides before photoacoustic detection to avoid signal interference with the tissues’ inherent signals. Frozen sections for H&E staining are less preferred due to their inferior histopathological presentation compared to paraffin sections. Similarly, IHC cannot be performed with frozen sections. Therefore, paraffin sections were chosen for photoacoustic data acquisition and comparison with H&E and IHC. As the inherent property of tissue influences photoacoustic imaging and spectral results, the findings of this study can theoretically be applied to frozen sections.

The H&E staining method, employing hematoxylin to stain cell nuclei blue and eosin to color the cytoplasm pink, is essential for distinguishing between normal and malignant tissue types, as illustrated in Figs. 2(b), 2(d), 2(f), 2(h), and Figs. 3(b), 3(d), 3(f), 3(h). Clinically, H&E staining can be challenging for diagnosing early-stage or junctional malignant tumors, often requiring additional diagnostic evaluations such as IHC biomarkers to accurately distinguish tumor developmental stages or types. For instance, Ki67 was used to identify the progressive stages of OSCC. In contrast, P63 and CK5 proteins were used for its diagnosis, and CK7 and CK8 were used for diagnosing ACC (the antibodies to the relevant proteins were purchased from Solarbio, China). All animal experiments complied with international regulations to protect experimental animals. The research protocol was approved by the Institute of Radiation Medicine, Chinese Academy of Medical Sciences (Approval No. IRM-DWLL-2023249).

3.2. Photoacoustic image acquisition, reconstruction, and virtual staining

We conducted photoacoustic imaging using a 266 nm wavelength laser with a Pulse Repetition Frequency (PRF) of 25 kHz and a pulse energy of 500 nJ. The scanning step size was set to 1.0 µm. This technique captured three-dimensional images of 5,000 pixels (x) × 5,000 pixels (y) × 100 pixels (z) within 20 minutes, covering a field of view (FOV) of 5 × 5 mm². Given that the section thickness was below the depth resolution of our bUV-PAM system (∼7.5 µm) [34], we employed Maximum Amplitude Projection (MAP) along the depth direction for each A-line signal to reconstruct the photoacoustic images, as illustrated in Figs. 2(a), 2(c), 2(e), and 2(g) and Figs. 3(a), 3(c), 3(e), and 3(g). To enhance the visual representation of the photoacoustic images, we applied a virtual staining method based on the corresponding H&E staining images. Specifically, since cell nuclei appear blue in H&E images and exhibit strong photoacoustic signals, they are represented in blue in the photoacoustic images. Regions with weaker photoacoustic signals, such as cytoplasm, connective tissues, and other extracellular materials, typically stain pink in H&E images and are colored accordingly. This virtual staining method produces photoacoustic images that closely resemble their H&E staining counterparts, thereby facilitating subsequent comparisons.

3.3. Photoacoustic spectra acquisition and analysis

Before photoacoustic spectra acquisition, it is imperative to perform an accurate pathological diagnosis to accurately align each spectral curve with its corresponding disease type. To achieve this, we employed traditional H&E staining and IHC biomarker methods for disease diagnosis, enabling the effective classification and labeling of different tissue areas. Leveraging the structural similarity between photoacoustic and H&E staining images, we delineated normal, OSCC, and ACC tissue regions within the respective unstained tissue sections for photoacoustic spectral acquisition (Figs. 2–5). Additionally, we used IHC images to characterize OSCC tissues at various stages (Fig. 6) and differentiate the types of OSCC and ACC tissues (Fig. 7). Using a two-dimensional motorized scanning stage with a step size of 20 µm, we positioned the samples precisely in targeted regions to capture disease-specific photoacoustic signals.

To mitigate potential laser-induced tissue damage, we reduced the excitation energy to 50 nJ at a repetition frequency of 15 kHz and employed a 512-time averaging mode to minimize noise in the photoacoustic signals. For each disease type, we collected 10 samples, obtaining 2000 photoacoustic A-line signals. We filtered the data, retaining the top 10% of the signals with the highest amplitude to ensure that the selected signals originated exclusively from the nuclei. Subsequently, we performed Fourier transformation of these time-resolved photoacoustic signals to extract the photoacoustic spectrum. By averaging all photoacoustic spectra curves, representative photoacoustic spectral curves for specific diseases were derived. The results of the experiments were analyzed using unpaired t-tests between two groups, while comparisons among three or more groups were performed using one-way ANOVA (GraphPad Prism 9.0, USA). Results were presented as mean value ± standard deviation, P < 0.05 indicated statistical significance.

4. Discussion and conclusion

Accurately diagnosing early-stage oral cancer and identifying specific cancer types are critical for effective clinical management. Histopathological evaluations rely heavily on the morphological features observed in H&E staining images to classify the pathological state of oral tumors, thereby offering crucial insights for clinical interventions. H&E staining offers specific nuclear morphological imaging contrast, resulting in inadequate diagnostic accuracy in early-stage cancer detection and the accurate classification of oral cancer types. IHC enables the identification of this disease through specific antibody biomarkers, but typically requires additional tissue samples, time, and expenses. IHC procedures often fail to meet these urgent demands during critical intraoperative situations that demand rapid pathological assessments.

In this study, we developed a novel diagnostic method for diverse oral cancer types by integrating PAMI with PASA. Leveraging our bUV-PAM system, our method allows simultaneous photoacoustic images and spectra data acquisition within 20 minutes (Fig. 1). Our PAMI technique provides label-free imaging of normal mucosa, parotid tissues, and their cancerous counterparts, with results corresponding to H&E staining images (Figs. 2 and 3). By analyzing the distinct variations in the photoacoustic spectra of different samples, our broadband PASA technology effectively distinguishes between normal mucosa tissue and OSCC, as well as between normal parotid tissue and ACC (Figs. 4 and 5). Although diagnosing progressive stages of OSCC and distinguishing between OSCC and ACC can be challenging with H&E staining images, our PASA technique addresses this issue efficiently, yielding corresponding pathological diagnostic outcomes in line with IHC results (Figs. 6 and 7).

While our proposed oral cancer diagnostic method offers distinct advantages, it has certain limitations. The signal-to-noise ratio (SNR) and resolution of photoacoustic images currently do not match those of H&E staining images. This restricts their utility in pathological diagnosis, requiring detailed observation of individual cell nuclei’ size and shape. Achieving better SNR in photoacoustic imaging demands enhanced acoustic sensitivity while maintaining a broad bandwidth, a goal dependent on cutting-edge optical ultrasound sensor developments [43]. The lateral resolution of the photoacoustic pathological imaging system is primarily determined by the NA of the ultraviolet objectives. In a transmission-mode OR-PAM system, a high NA ultraviolet objective significantly enhances lateral resolution [24]. However, this may pose challenges when imaging thicker tissues or conducting in vivo diagnosis as photoacoustic waves encounter difficulties in penetrating these samples. In contrast, a reflection-mode OR-PAM system excels at capturing reflected ultrasound signals, making it particularly effective for imaging thick samples. However, traditional reflection-mode OR-PAM systems often utilize large-sized ultrasound transducers that are opaque to ultraviolet lasers, thereby complicating their integration with high NA objectives. Thus, developing compact or ultraviolet-transparent ultrasound detectors allows for simultaneous realization of reflection-mode photoacoustic detection and high lateral resolution [44]. Furthermore, achieving imaging resolution also necessitates reducing the scanning step size, which in turn requires increase the pulse repetition frequency (PRF) of the excitation laser and the implementation of an efficient scanning mechanism to ensure that imaging duration remains unchanged [45].

The use of pseudo-color visualization for photoacoustic pathological images presents challenges, particularly concerning threshold selection and coloring techniques. Deep learning approaches offer the potential for precise conversions of photoacoustic images into H&E-liked representations, thus improving the accuracy of pathological diagnosis [46–48]. Moreover, the reliability of artificial intelligence (AI) in predicting IHC results is significantly influenced by the correlation between original images and the corresponding IHC images. Traditional approaches primarily analyze H&E staining images, focusing on the size, morphology, and distribution of nuclei, but often fail to clarify the relationships with protein expression evident in IHC images [49,50]. In contrast, our bUV-PAM system captures label-free morphological images similar to H&E staining while simultaneously acquiring photoacoustic spectral data directly correlated with IHC results. In addition, this integrated approach allows for the simultaneous capture of photoacoustic data and IHC biomarkers images on the same tissue section, thereby greatly facilitating image registration. Consequently, the combined use of photoacoustic imaging and broadband spectral analysis techniques holds great potential for enhancing the accuracy of IHC result predictions.

Current intraoperative margin pathology diagnostic methods, constrained by axial resolution, generate photoacoustic images that overlay nuclei at varying depths, inadequately representing the monolayer nuclei seen in H&E staining images [51,52]. Our bUV-PAM system achieves an outstanding axial resolution of approximately 7.5 µm, closely matching the typical thickness of tissue slices, generally ranging from 5-7 µm. Therefore, we anticipate that our photoacoustic tomographic images of fresh samples will closely resemble H&E staining images obtained through continuous mechanical sectioning. Moreover, by optimizing the focal depth of ultraviolet laser, we facilitate high lateral resolution imaging of fresh tissue samples with uneven surfaces [53]. Consequently, our approach has the potential to produce pathological images of single-layer nuclei at the intraoperative margin that closely resemble H&E staining images, thus enhancing the accuracy of pathological diagnosis.

Existing biomedical imaging techniques hold significant promise for the pathological diagnosis of oral cancer. Two-photon fluorescence microscopy (TPFM) enables the acquisition of three-dimensional images of fresh tissues with sub-micrometer lateral resolution and micrometer axial resolution within tens of minutes, thereby enhancing the efficiency of rapid tissue pathological diagnosis [54–56]. Similarly, Stimulated Raman Scattering (SRS) microscopy utilizes the strengths of Raman spectroscopy to facilitate rapid, label-free imaging of fresh tissue, boasting remarkable sensitivity and high chemical specificity [57–59]. While these optical imaging techniques provide intraoperative tissue images resembling H&E staining, they encounter difficulties in diagnosing oral cancer cases that elude detection by H&E staining. Our approach seamlessly integrates PAMI with PASA, achieving a high concordance with H&E staining images while also allowing for the assessment of tissue physical properties of through PASA. This integration leads to pathological diagnostic outcomes closely associated with those obtain through IHC, thereby offering distinct advantages in the pathological diagnosis of oral cancer.

In conclusion, we present an innovative method for diagnosing various types of oral cancer by integrating PAMI and broad-bandwidth PASA. This approach allows for the simultaneous acquisition of photoacoustic images and spectral curves within a single two-dimensional raster scan, significantly enhancing the diagnostic capabilities for different stages of tumor progression and distinguishing specific oral cancer types. This novel pathological diagnostic technique is a valuable complement to traditional H&E staining or IHC biomarker methods, especially in improving the early diagnosis and classification of oral cancers.

Funding

Guangdong Major Project of Basic Research ( 2020B0301030009); National Natural Science Foundation of China 10.13039/501100001809 ( 62405289, 12074203, 12174204); The Science and Technology Innovation Commission of Shenzhen ( JCYJ20220818101417039;); Shenzhen Peacock Plan 10.13039/501100012234 ( KQTD20170330110444030); Scientific Instrument Developing Project of Shenzhen University ( 2023YQ001); Shenzhen University 2035 Initiative ( 2023B004); Key R&D Program of Zhejiang ( 3000-3AA240100); Tianjin Key Medical Discipline (Specialty) Construction Project ( TJYXZDXK-009A); Medical Talent Project of Tianjin City ( TJSJMYXYC-D2-032); Medical Education Research Project of Medical Education Branch of Chinese Medical Association ( 2023B175).

Disclosures

The authors declare that there are no conflicts of interest.

Data availability

Data underlying the results presented in this paper is not publicly available at this time but may be obtained from the corresponding authors upon reasonable request.