Enhanced tissue compartmentalization by applying segmentation algorithms on reduced spatial omics data

Significance of the topic

The ability to accurately segment spatial omics data from MALDI imaging is critical for revealing tissue heterogeneity and underlying pathological features. Traditional clustering on high-dimensional pixel spectra can be computationally intensive and susceptible to noise. Incorporating dimensionality reduction prior to segmentation offers a promising route to improve speed, clarity and the biological relevance of cluster maps.

Objectives and study overview

This study evaluates how combining component analysis (CA) with advanced segmentation algorithms enhances tissue compartmentalization in reduced spatial omics datasets. Using both MALDI Imaging and MALDI HiPLEX-IHC data from rat kidney and human lung carcinoma, the work compares classic k-means clustering on raw intensities with segmentation on CA scores and modern workflows based on UMAP dimension reduction and Leiden clustering. Performance is benchmarked in SCiLS Lab 2026a.

Methodology and instrumentation

Data from MALDI Imaging and HiPLEX-IHC experiments were imported into SCiLS Lab 2026a. For MALDI HiPLEX-IHC, a predefined target list extracted antibody-derived mass tags, while T-ReX® feature finding detected 150 lipid features in rat kidney. Feature intensities were normalized to total ion count (TIC). Dimensionality reduction approaches included principal component analysis (PCA) and uniform manifold approximation and projection (UMAP). Segmentation methods comprised standard k-means (k=10) on raw and denoised intensities or PCA scores, as well as Leiden clustering applied to UMAP embeddings. Spatial denoising was optionally applied to feature images or component score maps to minimize noise from sparse low-intensity pixels.

Main results and discussion

Clustering directly on raw intensities yielded noisy, fragmented regions that obscured morphological structures. Applying weak spatial denoising to intensities improved cluster coherence marginally. Segmentation on PCA component scores dramatically reduced noise while preserving fine tissue detail. The strongest performance emerged from combining UMAP reduction with Leiden clustering, which captured non-linear relationships in the data and generated crisp, biologically meaningful compartments such as glomeruli in kidney and tumor boundaries in lung carcinoma. Spatial smoothing of component maps further enhanced the visual continuity of clusters.

Benefits and practical applications of the method

Integrating CA or UMAP prior to segmentation greatly reduces data dimensionality and noise, accelerating computational workflows and revealing subtle biological patterns. Laboratories analyzing large MALDI-based spatial omics datasets can adopt these techniques to produce high-resolution tissue maps for disease research, biomarker discovery and quality control in pharmaceutical development.

Used instrumentation

• MALDI Imaging mass spectrometry systems
• MALDI HiPLEX-IHC multiplexed antibody reagent platform
• SCiLS Lab 2026a software with multivariate statistics module
• T-ReX® feature finding algorithm
• UMAP and Leiden clustering algorithms

Future trends and potential applications

Emerging dimensionality reduction tools and graph-based clustering methods will further refine spatial omics segmentation. Integration with deep learning for automated feature extraction and real-time processing pipelines may enable interactive mapping of clinical samples. Combining multimodal imaging data promises richer molecular context for single-cell and subcellular resolution studies.

Conclusion

This work demonstrates that performing segmentation on reduced feature spaces, particularly UMAP followed by Leiden clustering, outperforms traditional k-means on raw data. The approach enhances computational efficiency, suppresses noise and produces biologically coherent tissue compartments. These workflows, available in SCiLS Lab 2026a, provide a robust framework for advanced MALDI imaging analysis.