Accurate assessment of surgical margins remains a central challenge in breast-conserving surgery, where residual tumour at the resection edge can lead to repeat operations and worse outcomes. Existing intraoperative techniques can reduce re-excision rates but often depend on time-consuming pathology workflows, specialist expertise or limited molecular specificity. Against this backdrop, interest has grown in approaches that combine rapid morphological imaging with molecular information capable of distinguishing malignant from tumour-adjacent noncancerous breast tissue. A multimodal strategy integrating femtosecond label-free imaging microscopy with mass spectrometry imaging and machine learning has been evaluated to address this gap. The approach seeks to deliver rapid, label-free visualisation of tissue architecture alongside metabolic signatures, with the aim of improving margin assessment during breast-conserving procedures and supporting more precise surgical decision-making.
Label-Free Microscopy for Rapid Morphological Delineation
Femtosecond label-free imaging microscopy was assessed for its ability to visualise breast tissue morphology without sectioning or staining. The technique captures autofluorescence, second harmonic generation and third harmonic generation signals in a single scan, enabling depiction of epithelial structures, collagen-rich stroma and lipid interfaces. Imaging was performed on both freshly resected tissue blocks and paraformaldehyde-fixed specimens, demonstrating consistent image quality across preservation methods. Cellular features observed with this approach closely matched those seen on standard haematoxylin and eosin-stained sections, including distinctions between tumour regions and tumour-adjacent noncancerous breast tissue.
The method showed sufficient penetration depth to visualise tumour cells embedded within surrounding adipose tissue, a context where conventional thin-section histology may not fully capture spatial relationships relevant to margin assessment. Imaging speed was also notable, with large fields of view acquired within seconds and comprehensive assessment completed within minutes. These characteristics support the feasibility of integrating femtosecond label-free imaging into intraoperative or near-operative workflows, where timely feedback on margin status is essential.
Molecular Profiling Through Mass Spectrometry and Machine Learning
While label-free microscopy provides detailed morphological information, it does not reveal molecular composition. To address this limitation, matrix-assisted laser desorption ionisation mass spectrometry imaging was applied to tissue sections corresponding to imaged regions. This enabled spatial mapping of metabolites and lipids without labelling, producing molecular segmentation maps that aligned with histological features identified by microscopy.
Machine learning models were trained on metabolomic features extracted from tumour and tumour-adjacent noncancerous breast tissues. Several algorithms were evaluated, including random forest, decision tree, support vector machine and extreme gradient boosting. All achieved balanced classification accuracy of at least 80%, with extreme gradient boosting delivering the strongest overall performance at both micro-region and patient levels. Feature importance analysis across models converged on a small panel of discriminating metabolites, notably taurine, threonate and glutamate.
These metabolites were consistently more abundant in breast cancer tissue than in adjacent noncancerous tissue. When combined in a reduced model, they achieved high discriminative performance, which was maintained in an independent external dataset. Among the three, taurine emerged as the most prominent feature, showing high abundance, strong classification metrics and consistent elevation across cohorts.
Must read: Medical VLMs Falter Under Common Imaging Artefacts
Taurine as a Marker of Margin Status and Biological Relevance
The potential clinical relevance of taurine was further explored in the context of surgical margin assessment. Analysis of tissue sections containing defined tumour cores and graded margin distances showed a clear spatial gradient, with taurine intensity highest in tumour regions and progressively decreasing with increasing distance from the tumour edge. This pattern suggests utility in distinguishing positive, close and negative margins based on metabolic signal intensity.
Beyond spatial discrimination, taurine levels were examined in relation to patient outcomes using published metabolomic and prognostic data. Higher taurine abundance was associated with poorer overall survival among patients with breast cancer. Complementary functional experiments in breast cancer cell lines demonstrated that taurine exposure promoted cell viability, proliferation and cell cycle progression while reducing apoptosis. These findings support a pro-tumorigenic role for taurine and provide biological context for its association with adverse prognosis.
Taken together, the imaging, computational and functional data position taurine as a candidate biomarker that links metabolic activity with both local margin status and broader disease behaviour. Its detectability through mass spectrometry imaging and its spatial correlation with tumour boundaries underscore its potential value in surgical settings.
The integration of femtosecond label-free imaging microscopy with mass spectrometry imaging and machine learning offers a combined morphological and molecular framework for breast cancer margin assessment. The approach delivers rapid, label-free visualisation comparable to conventional histology while adding metabolic information that enhances tissue discrimination. Within this framework, taurine has been identified as a prominent marker associated with tumour presence, margin proximity and patient outcomes. These findings highlight a pathway toward more precise intraoperative assessment, with the potential to reduce re-excision rates and support tailored surgical management without reliance on lengthy pathological workflows.
Source: npj digital medicine
Image Credit: iStock
