Integrating Spatial and Morphological Characteristics into Melanoma Prognosis: A Computational Approach

Abstract

In this study, the prognostic value of cellular morphology and spatial configurations in melanoma has been examined, aiming to complement traditional prognostic indicators like mitotic activity and tumor thickness. Through a computational pipeline using machine learning and deep learning methods, we quantified nuclei sizes within different spatial regions and analyzed their prognostic significance using univariate and multivariate Cox models. Nuclei sizes in the invasive band demonstrated a significant hazard ratio (HR) of 1.1 (95% CI: 1.03, 1.18). Similarly, the nuclei sizes of tumor cells and Ki67 S100 co-positive cells in the invasive band achieved HRs of 1.07 (95% CI: 1.02, 1.13) and 1.09 (95% CI: 1.04, 1.16), respectively. Our findings reveal that nuclei sizes, particularly in the invasive band, are potentially prognostic factors. Correlation analyses further demonstrated a meaningful relationship between cellular morphology and tumor progression, notably showing that nuclei size within the invasive band correlates substantially with tumor thickness. These results suggest the potential of integrating spatial and morphological analyses into melanoma prognostication.

Keywords: cellular morphology; computational pipeline; deep learning; machine learning; melanoma prognostication; spatial analysis.

Figure 1

Multiplex Immunohistochemistry (mIHC) Visualization from Our Cohort. The image showcases a melanoma tissue section stained with DAPI (red) for nuclei, Ki67 (green) indicating proliferating cells, and S100 (blue) highlighting melanoma cells. The inset provides a magnified view of the indicated region, demonstrating the cellular detail captured by mIHC.

Figure 2

Overview of the computational analysis pipeline for melanoma prognosis: Step1. Acquire the mIHC image slides and annotation of the melanoma site and epidermis. Step2. Detect cells within the melanoma region. Step3. Measure cellular features. Step4. Threshold cells based on the marker intensity to distinguish positive cell counts of different markers. Step5. Quantify spatial and morphological features within different tumor bands. Step6. Conduct survival analysis based on spatial, morphological, and clinical information.

Figure 3

The quality control process of the dataset.

Figure 4

Histogram of the marker expression data overlaid with the BGMM thresholding process, as determined by the EM simulation. The yellow bars represent the $T_1$ distribution, the red curve indicates the total fit of the BGMM, and the blue curve illustrates the individual Gaussian components. The intersection of these components, marked by the dashed green line, defines the optimal threshold for segmenting positive cells from negative ones.

Figure 5

Sample visualization of the spatial analysis process. Panel (A) displays a fluorescence microscopy image of the melanoma sample, highlighting S100+ (blue) and Ki67+ (green), and DAPI (red). Panel (B) shows a schematic of the spatial analysis results: the melanoma region is segmented into three distinct spatial bands: the Superficial Band (white), Middle Band (blue), and invasive band (dark blue). The thickness of each band is derived from PCA-based tumor thickness measurements, indicated by the dashed magenta line.

Figure 6

Forest plot of concordance indices for prognostic factors in melanoma. This plot illustrates the concordance indices and their 95% confidence intervals for various prognostic factors, including cell density and nuclei sizes across different tumor bands, as well as traditional factors such as age, mitosis rate, and tumor thickness.

Figure 7

Boxplots of cell nuclei sizes across different tumor bands. The boxplots compare the distribution of cell nuclei sizes within the invasive, middle, and superficial bands of melanoma tumors, including mean cell nuclei size, mean tumor cell nuclei size, and mean colocalized (Ki67+ S100+) cell nuclei size.

Figure 8

Composite correlation heatmaps of cell, tumor cell, and colocalized cell nuclei sizes with traditional prognostic factors in melanoma. Each subfigure represents the pairwise correlation coefficients between nuclei sizes in different tumor bands and standard prognostic measures, such as mitosis rate and tumor thickness, across three distinct cellular categorizations: (a) cell nuclei size, (b) tumor cell nuclei size, (c) colocalized cell nuclei size. (Asterisks denote levels of statistical significance: * p < 0.05, ** p < 0.01, *** p < 0.001).