1 Human cSCC manifestation in Pathology
The pathological examination revealed distinct pathological alterations associated with cSCC, which originates from keratinocytes. It can be categorized into three subtypes based on the degree of keratinization: well differentiated, moderately differentiated, and poorly differentiated. In this study, our focus is primarily on well differentiated and poorly differentiated cases. In well differentiated cSCC, the tumor cells exhibit squamous morphology, with evident intercellular bridges. Mitotic figures are infrequent, and there is a presence of cells displaying poor keratinization, as well as keratin pearls. Conversely, poorly differentiation is characterized by a disorganized arrangement of cells, a significant number of atypical cells and mitotic figures, and either an absence or rarity of keratinization. Figure 1 exhibits the pathological image of normal skin and cSCC.
2 Features of human cSCC were highlighted by MSI system.
To minimize discrepancies arising from conventional staining methods, this study utilized MSI system equipped with LED illumination, encompassing 13 wavelengths, with five images per histological sample. This approach enables the utilization of spectral dimensions for precise representation of morphological characteristics specific to different tissue types.
Figure 2 shows the distinctive structural features of normal skin and cSCC in single-spectral band and pseudo-color composite images. The images were segmented by adaptive threshold segmentation method at 520nm and 600 nm spectral band and random forest segmentation method, respectively. Specifically, the images at 520nm spectral band were used for the segmentation of collagen fibers in normal tissue and diseased epidermis in cancerous tissue. The image at 600 nm spectral bands is used to segment large lipid droplets in normal tissues and keratin pearls in cancerous tissues. The images at 630nm spectral band were primarily used to segment the cellular nucleus. Keratosis was more obvious in well differentiated cSCC compared to poorly differentiated ones. The mixed image of pseudo-color not only qualitatively distinguishes normal skin, poorly and well differentiated tissue, but also mitigates the impact of staining differences on diagnostics, offering pathologists a more objective and accurate diagnostic basis.
3 Differentiation between normal skin and cSCC was accomplished through nucleo-plasmic ratio.
High nucleo-plasmic ratio serves as a crucial indicator of tumor development and carcinogenesis, for that nucleus density is elevated in cancer cells. To conduct a more precise analysis, we adopted the sliding window technique to create a 200x200 window to focus specifically areas. The selection was confined by a threshold value to exclude regions with underexposed images or where the boundary between the nucleus and keratin pearls was unclear. Subsequently, the nucleo-plasmic ratio was calculated within the automatically selected five windows.
As shown in Fig. 3A, the mosaic images depict normal, well and poorly differentiated tissues. The nucleo-plasmic ratio of normal skin is significantly lower compared to that of cSCC. Statistical analysis was performed on ten windows for each slice, and the distribution of the nucleo-cytoplasmic ratio is shown in Fig. 3B. which also shows that there is a significant difference in nuclei area between normal skin and cSCC (P < 0.001) (Fig. 3C). For further discrimination between these tissues, we conducted an in-depth analysis of the specificity and accuracy of the nuclear area proportion. The resultant ROC curve comparisons among normal, poorly differentiated, and high-differentiated tissues are displayed in Fig. 3D.
4 Poorly and well differentiated cSCC was identified via nuclear atypia and keratin pearls.
Although it is possible to differentiate cSCC from normal skin based on nucleo-plasmic ratio, this method does not provide discrimination between high and poor levels of differentiation. Therefore, the examination of keratin pearls and the quantitative analysis of nuclear heterogeneity were investigated.
The nuclear size was calculated within segmented individual nuclear images as a measure of nuclear heterogeneity. The proportion of nuclei below 10um2, 20um2, 30um2, and 40um2 relative to the total number of nuclei in the segmentation image were calculated from five fields of section on each slide. The analysis revealed that utilizing a threshold of 20um2 effectively distinguished with poorly and well differentiated cSCC. All nuclei exceeding 20um2 were distinctly marked on the segmentation image (Fig. 4A). In the subsequent data analysis, five images were evaluated for each specimen slide. We calculated the proportion of cell nuclei having an area smaller than 20µm² in each image. The percentage of nuclei smaller than 20um2 in poorly differentiated cSCC was significantly lower than that in well differentiated tissues (P < 0.001) (Fig. 4B,4C). Then, we conducted specificity and accuracy analyses to thoroughly evaluate the number and proportion of nuclei, and the results of the ROC curve is shown in Fig. 4D.
In this study, we also found poorly and well differentiated cSCC could be distinguished by keratin pearls. By quantifying ten non-contiguous regions on each slide, the results indicated that in well differentiated cSCC, the area of keratin pearls was significantly larger and more abundant compared to poorly differentiated samples. Figure 5A,5B presents the relevant pseudo-color diagram depicting segmentation. Figure 5C presents the quantity of keratin pearls within ten windows for each specimen slide. Slides 1 through 5 are categorized as poorly differentiated, while slides 6 through 10 are classified as well differentiated.