Large histological serial sections for computational tissue volume reconstruction.

Objectives: A proof of principle study was conducted for microscopic tissue volume reconstructions using a new image processing chain operating on alternately stained large histological serial sections.

Methods: Digital histological images were obtained from conventional brightfield transmitted light microscopy. A powerful nonparametric nonlinear optical flow-based registration approach was used. In order to apply a simple but computationally feasible sum-of-squared-differences similarity measure even in case of differing histological stainings, a new consistent tissue segmentation procedure was placed upstream.

Results: Two reconstructions from uterine cervix carcinoma specimen were accomplished, one alternately stained with p16(INK4a) (surrogate tumor marker) and H&E (routine reference), and another with three different alternate stainings, H&E, p16(INK4a), and CD3 (a T-lymphocyte marker). For both cases, due to our segmentation-based reference-free nonlinear registration procedure, resulting tissue reconstructions exhibit utmost smooth image-to-image transitions without impairing warpings.

Conclusions: Our combination of modern nonparametric nonlinear registration and consistent tissue segmentation has turned out to provide a superior tissue reconstruction quality.