Figure 1 summarizes the overall study design and pathological evaluation workflow. To investigate the diagnostic potential of OCT across a broad spectrum of thyroid carcinoma, OCT imaging and quantitative analysis were performed on surgically resected thyroid cancer specimens from 41 patients, including PTC (n = 11), FTC (n = 6), OCA (n = 3), ATC (n = 9), and MTC (n = 12). Representative tissue sections for OCT imaging were selected by a pathologist, and paraffin blocks were reciprocally processed to obtain formalin-fixed specimens suitable for direct comparison with histologic sections.

Fig. 1

Overview of the OCT-based pathological study for thyroid carcinoma. A schematic diagram presents the overall workflow of the OCT-based pathological analysis in thyroid carcinoma. Resected human thyroid specimens, fixed in formalin, were imaged employing a benchtop swept-source OCT system. Three-dimensional OCT datasets were acquired from pre-selected regions of interests (ROIs), and en face OCT images were subsequently reconstructed. These en face images were co-registered with corresponding histological sections. Representative ROIs from each case were then selected for detailed pathological interpretation and quantitative analysis

OCT imaging was performed using a benchtop swept-source (SS)-OCT system, enabling acquisition of volumetric datasets for en face visualization and 3D reconstruction of tissue architecture. Through direct comparison with corresponding histologic sections, OCT findings were used to identify characteristic architectural and compositional signatures of thyroid carcinoma and to evaluate depth resolved morphological changes using the volumetric imaging capability of OCT. For quantitative validation of OCT based discrimination across the broad spectrum of thyroid carcinoma, we evaluated whether OCT images capture measurable differences between NT and carcinoma, as well as among distinct carcinoma subtypes. OCT derived parameters reflecting tissue morphological characteristics, including surface irregularity, penetration depth, and tissue brightness, were defined and analyzed. In addition, texture features derived from GLCM analysis were evaluated to capture cellular level differences within tissue microstructure across thyroid carcinoma subtypes. Detailed descriptions are provided in the “Methods” section.

Fig. 2

Representative OCT images of each thyroid carcinoma group. The B-scan, en face OCT image, and the corresponding histological section are displayed. The position of the B-scan within the en face plane is indicated at the top left of each B-scan image and adjacent to the corresponding en face image. a PTC. The PTC shows compact papillary growth with slit-like space (arrow) in en face OCT image. The collagen fiber of tumor fibrous capsule shows its own alignment direction (asterisk). b FTC. The intra-tumoral follicular structure is evident in FTC except the microfollicular area (arrow). The elongated collagen fiber of tumor fibrous capsule is identified with multilayering appearance (asterisk). c OCA. The microfollicular pattern of OCA is identified as a relatively homogeneous solid feature with high signal intensity in OCT images. d ATC. The infiltrative feature of ATC is demonstrated replacing normal follicular structure. The signal intensity is relatively low and the penetration depth is short comparing to the adjacent normal thyroid parenchyma. e MTC. Diverse histologic pattern in MTC is identified (arrow: papillary, asterisk: conventional). All scale bars, 1 mm

2.1 Qualitative analysis results: en face OCT image correlation with H&E pathologic images

Representative OCT images of the five thyroid carcinoma groups, namely PTC, FTC, OCA, ATC, and MTC, are presented in Fig. 2. Each case includes B-scan images, en face reconstructions, and corresponding histological sections to enable direct morphological correlation.

2.1.1 PTC

In en face OCT of conventional PTC, the tumor showed bright reflective signal and elongated papillary structure with slit-like spaces in a vaguely solid mass (Fig. 2a). When follicular pattern existed, the colloid within intra-tumoral follicles resembled normal thyroid colloid (Fig. 2a). The fibrous capsule showed collagen fibers oriented perpendicular to tumor growth, and the brightness varied with tissue density (Fig. 2a). Capsular invasion was identified by a bright tumor bud disrupting the moderately dark fibrous capsule. The psammoma bodies, lamellated small round calcification, appeared as highly bright round spots (Supplementary Fig. 1a). When densely packed, papillary architecture was indistinct, appearing as internal coarseness within solid mass (Supplementary Fig. 1b). Solid/trabecular PTC showed relatively homogeneous solid sheet separated by thin fibrous septa (Supplementary Fig. 1c). In tall cell PTC, elongated papillae with tram-track appearance were identified (Supplementary Fig. 1d). Hobnail features were not clearly demonstrated, likely due to resolution limits.

2.1.2 FTC

In Fig. 2b, the B-scan OCT image showed FTC mass filled with intra-tumoral follicles with fibrous capsule. The pushing effect of tumor is evident in the en face OCT image, flattened follicles in the adjacent thyroid parenchyma pressed by the mass. Normo- or macro-follicular patterns were readily identified, while microfollicular pattern appeared solid in OCT. The intra-tumoral follicular structure was well delineated in terms of shape, size, density, and growth direction. The tumor cells were relatively bright compared to the adjacent NT parenchyma (Fig. 2b). Thick fibrous capsules appeared multilayered (Fig. 2b). Calcified stroma was highly reflective.

2.1.3 OCA

The OCA was composed of oncocytic cells, having abundant eosinophilic granular cytoplasm filled with mitochondria. In this cohort, the OCA showed mainly microfollicular pattern growth. The architectural features were similar to FTC (Fig. 2c), but the tumor cells exhibited relatively higher signal intensity compared to those of FTC (Fig. 2c).

2.1.4 ATC

The ATC showed highly infiltrative tumor border (Fig. 2d). The undifferentiated tumor cells exhibited a pattern-less growth pattern, frequently admixed with inflammatory cells. Giant cell-rich ATC showed multiple infiltrative foci with coarse granularity (Fig. 2d). Squamous carcinoma pattern showed irregular sheets of tumor cells with brighter reflective signal than dark inflammatory cell-rich stroma (Supplementary Fig. 2a). The geographic necrosis was frequently observed in ATC, where the reflective signal was low and appeared slightly dark (Supplementary Fig. 2b).

Fig. 3

Depth-dependent structural variations visualized by OCT imaging. The depth-specific architectural alteration is revealed by OCT imaging. The corresponding histological section is shown alongside the en face OCT image. The image on the left corresponds to the original en face plane, with subsequent frames arranged toward the right with increasing depth at same intervals. a Papillary architecture in PTC. Note the retained high signal intensity of the outer layer of papillary structure, in which the carcinoma cells align, until greater depths. b The capsular invasion focus of FTC. The intra-tumoral follicular structure appears elongated perpendicular to the fibrous tumor capsule representing the pushing force vector (arrows). The intracapsular vascular space is progressively occupied by fibrinous material (asterisk). c The OCT images at successive depth intervals demonstrate decreasing tumor dimension in this medullary thyroid carcinoma case. This indicates the three-dimensional ovoid nature. The red line indicates the tumor outline at each imaging plane, with the red dotted line representing the outline from the first plane for comparison. Tumor boundaries were manually delineated with reference to the corresponding histological sections and reviewed by a certified pathologist (JYS). All scale bars, 1 mm

2.1.5 MTC

The diverse architectural pattern of MTC was well identified; conventional, papillary, solid, and cystic (Fig. 2e and Supplementary Fig. 2c). The stromal amyloid deposition area showed relatively brighter appearance with high reflective signal (Supplementary Fig. 2d).

2.1.6 Non-neoplastic tissues

The follicular epithelial cells appeared as thin rims of round structures with moderately bright signals (Supplementary Fig. 3a). The intra-follicular colloid showed a dark homogeneous texture, with brightness varying according to density (Supplementary Fig. 3a). Follicular nodular disease displayed hyperplastic follicles forming nodules, with variable size and shape—micro-, normo-, macro-follicular and occasional cystic change (Supplementary Fig. 3b).

In lymphocytic thyroiditis, lymphoid follicles appeared granular, round to ovoid with fuzzy outlines (Supplementary Fig. 3c). Mild infiltration and fibrosis caused parenchymal haziness with follicular remnants (Supplementary Fig. 3b).

The normal parathyroid parenchyma was slightly bright with occasional dark vacuoles (Supplementary Fig. 4a). At this resolution, detailed architecture was indistinct.

The lymph node showed bright capsules and distinct lymphoid follicles; metastatic PTC displayed intensely bright psammoma bodies (Supplementary Fig. 4b).

The fat had a honeycomb appearance with dark vacuoles (Supplementary Fig. 4c). The arteries were round to ovoid with multilayered signals; smooth muscle was relatively bright. Blood often appeared bright. Skeletal muscle fascicles were outlined by dark connective tissue (Supplementary Fig. 4d).

2.2 Qualitative analysis results: 3D reconstruction image analysis2.2.1 Papillary architecture in PTC

Sequential en face images reveal gradual transitions in papillary architecture (Fig. 3a). At the 0 μm-depth level, the papillary projections were visualized within the intra-tumoral vacant spaces. Papillary outlines were lined with PTC cells, showing linear bright reflective signal from 67 to 134 μm-depth level. The central fibro-vascular core appeared relatively dark compared to the outlining tumor cells, which remained persistently dark and showed gradually decreasing signal intensity (Supplementary Movie 1).

2.2.2 Capsular invasion and peri-capsular vascular structure in FTC

At the capsular invasion focus of FTC, the neoplastic follicles were elongated along the invasion vector and oriented perpendicularly to the fibrous capsule (Supplementary Movie 2). The peri-capsular vascular space was evident across depths, revealing intravascular contour changes and highly reflective fibrin along the intima (Fig. 3b).

2.2.3 Tumor diameter change in MTC

Changes in tumor contour and size across different depths are visualized and measured (Fig. 3c). The maximal tumor diameter gradually decreased from a depth level of 101 μm to 202 μm, indicating a spherical 3D shape. OCT, which provides structural 3D morphology, enabled analysis of the geometric outline of the tumor across imaging depths regardless of tissue optical properties or OCT signal attenuation.

Fig. 4

Quantitative analysis results. The surface irregularity, penetration depth, tissue brightness and GLCM texture analysis shows discriminative power in various thyroid carcinoma groups. The surface irregularity is high in PTC and FTC comparing to other groups, in which papillary or follicular architecture is predominant. The penetration depth in FTC is notably high comparing to other groups. The tissue brightness is relatively high in PTC, FTC and OCA comparing to ATC or NT. The GLCM texture analysis provides additional discriminative power. Penetration depth was evaluated at the A-scan level using randomly selected measurements with equal sample size for each carcinoma group (n = 10,000 per group), whereas for all other metrics n corresponds to the number of cases in each group (see “Methods” for details). Data are presented as Tukey box plots

2.3 Quantitative analysis results2.3.1 Surface irregularity

Carcinomas exhibited flatter surface profiles than NT (P < 0.0001) (Table 1), reflecting increased intra-tumoral compact cellular growth and stromal remodeling that replaces the normal thyroid follicular structure.

Table 1 Summary statistics of quantitative metrics comparing carcinoma and normal thyroid gland tissue

Intergroup comparisons revealed statistically significant differences between each carcinoma group and NT, except for FTC (Fig. 4; Table 2). ATC, exhibiting the flattest surface feature, was clearly distinguishable from PTC, FTC, and MTC (P = 0.0007, 0.0008, and 0.0227). This likely reflects the solid growth and high cellularity of undifferentiated ATC cells lacking architectural differentiation. Notably, the higher values observed in PTC and FTC reflect increased surface irregularity associated with their respective papillary or follicular architectures. This was also evident in the comparison between ATC and the DTC as a whole (3.24 ± 2.49 μm; P = 0.0023).

2.3.2 Penetration depth

Penetration depth serves as an indirect indicator of tissue architecture because optical scattering and attenuation vary with the cellular density and stromal composition characteristic of each pathological state. Carcinomas showed significantly shorter penetration depth than NT (P < 0.0001) (Table 1). This reflects replacement of low-scattering follicles by densely packed tumor cells and fibrotic stroma.

Comparison of penetration depth between each thyroid carcinoma group and NT revealed statistically significant differences across all carcinoma groups relative to NT (Fig. 4; Table 2). FTC demonstrated the greatest penetration depth, showing a statistically significant distinction from all other carcinoma groups. This finding is likely attributable to its preserved macro- or normo-follicular architecture and the presence of colloid-filled follicles, which closely resemble the features of NT. Despite its architectural similarity to NT, FTC exhibited a slightly reduced penetration depth compared to NT. This could be explained by the presence of intermixed microfollicular patterns or stroma, which introduce moderate optical scattering.

Table 2 Cross-sectional comparison of quantitative metrics between normal thyroid gland tissue and carcinoma subtype2.3.3 Tissue brightness

A marked increase in tissue brightness was observed in carcinoma samples compared to NT (P < 0.0001) (Table 1). The elevated brightness likely reflects increased cellular density, loss of organized follicular architecture with loss of colloid, and extracellular-matrix remodeling. These structural and compositional changes are likely to influence local light-tissue interactions, contributing to the observed optical signal enhancement in carcinoma regions.

Each carcinoma group except ATC exhibited significantly higher tissue brightness compared to NT (Fig. 4; Table 2). Despite its solid and compact growth pattern, ATC demonstrated significantly lower brightness, which may reflect distinct histopathological features such as abundant intra-tumoral inflammatory infiltration and/or necrosis. This optical phenotype of ATC differed significantly from PTC, OCA, and MTC. DTC showed higher brightness than ATC (26.43 ± 2.30 dB; P = 0.0002).

2.3.4 GLCM texture analysis

All five GLCM-derived texture metrics (“contrast”, “entropy”, “correlation”, “homogeneity”, and “energy”) demonstrated statistically significant differences between carcinoma and NT (P < 0.0001) (Table 1). PTC, OCA, ATC and MTC were distinguishable from NT across all five GLCM-derived texture parameters (Fig. 4; Table 2). The “correlation” exhibited significant differences between FTC and NT, indicating that OCT-derived texture analysis can reveal subtle microstructural differences even when the overall histoarchitectural appearance appears similar. “Contrast”, “entropy” and “correlation”, PTC was distinguished from MTC, while other metrics could not provide the discriminative power. Notably, “contrast” also demonstrated the ability to differentiate between OCA and MTC, a distinction not achieved by any other texture features. “Contrast” and “entropy”, reflecting heterogeneity, clearly separated DTC (147.98 ± 26.30 a.u. and 7.07 ± 0.40 a.u.) from ATC and MTC.

2.4 Preliminary evaluation of diagnostic classification between NT and carcinoma

To explore the potential of the proposed OCT-derived metrics as quantitative biomarkers for pathological differentiation, a preliminary classification analysis was performed between NT and carcinoma groups. Receiver operating characteristic (ROC) analysis demonstrated that the evaluated OCT-derived metrics each showed discriminative potential for separating NT from carcinoma by capturing complementary quantitative characteristics, with the correlation metric exhibiting the highest overall performance among the parameters (Table 3 and Supplementary Fig. 5). These results suggest that the proposed quantitative descriptors have potential utility for pathological classification.

Table 3 Summary of the diagnostic performance of OCT-derived metrics for differentiating carcinoma from normal thyroid tissue