Search

Einenkel J, Braumann U-D, Steller W, Binder H & Horn L-C (2012) Histopathology 60, 1084–1098 Suitability of infrared microspectroscopic imaging for histopathology of the uterine cervix Aims: Infrared microspectroscopy (IR-MSP) has been proposed for automated histological tissue differentiation of unstained specimens based on chemical analysis of cell and extracellular constituents. This study aimed to determine the accuracy of IR-MSP-based histopathology of cervical carcinoma sections with complex tissue architecture under practically relevant testing conditions. Methods and results: In total, 46 regions of interest, covering an area of almost 50 mm2 on sections derived from paraffin-embedded tissue of radical hysterectomy specimens, were analysed by IR-MSP (nominal resolution ∼4.2 μm). More than 2.8 million pixel spectra that were processed using fuzzy c-means clustering followed by hierarchical cluster analysis permitted image segmentation regarding different biochemical properties. Linear image registration was applied to compare these segmentation results with manual labelling on haematoxylin and eosin-stained references (resolution ∼0.7 μm). For recognition of nine tissue types, sensitivities were 42–91% and specificities were 79–100%, mostly being affected by peritumoral inflammatory responses. Algorithmic variation of the outline of dysplasia and carcinoma revealed a spatial preference of false values in tissue transition areas. Conclusions: This imaging technique has potential as a new method for tissue characterization; however, the recognition accuracy does not justify a pathologist-independent tissue analysis, and the application is only possible in combination with concomitant conventional histopathology.

Infrared microspectroscopy (IR-MSP) has been proposed for automated histological tissue differentiation of unstained specimens based on chemical analysis of cell and extracellular constituents. This study aimed to determine the accuracy of IR-MSP-based histopathology of cervical carcinoma sections with complex tissue architecture under practically relevant testing conditions. In total, 46 regions of interest, covering an area of almost 50 mm(2) on sections derived from paraffin-embedded tissue of radical hysterectomy specimens, were analysed by IR-MSP (nominal resolution ~4.2 μm). More than 2.8 million pixel spectra that were processed using fuzzy c-means clustering followed by hierarchical cluster analysis permitted image segmentation regarding different biochemical properties. Linear image registration was applied to compare these segmentation results with manual labelling on haematoxylin and eosin-stained references (resolution ~0.7 μm). For recognition of nine tissue types, sensitivities were 42-91% and specificities were 79-100%, mostly being affected by peritumoral inflammatory responses. Algorithmic variation of the outline of dysplasia and carcinoma revealed a spatial preference of false values in tissue transition areas. This imaging technique has potential as a new method for tissue characterization; however, the recognition accuracy does not justify a pathologist-independent tissue analysis, and the application is only possible in combination with concomitant conventional histopathology.