Detection and segmentation of cell nuclei in virtual microscopy images: a minimum-model approach

Abstract

Automated image analysis of cells and tissues has been an active research field in medical informatics for decades but has recently attracted increased attention due to developments in computer and microscopy hardware and the awareness that scientific and diagnostic pathology require novel approaches to perform objective quantitative analyses of cellular and tissue specimens. Model-based approaches use a priori information on cell shape features to obtain the segmentation, which may introduce a bias favouring the detection of cell nuclei only with certain properties. In this study we present a novel contour-based "minimum-model" cell detection and segmentation approach that uses minimal a priori information and detects contours independent of their shape. This approach avoids a segmentation bias with respect to shape features and allows for an accurate segmentation (precision = 0.908; recall = 0.859; validation based on ∼8000 manually-labeled cells) of a broad spectrum of normal and disease-related morphological features without the requirement of prior training.

Figure 1

H&E stained tissue samples: (a) breast cancer; (b) liver; (c) gastric mucosa; (d) bone marrow with primary myelofibrosis with highly pleomorphic megakaryocytes (without cell cluster separation); (e) connective tissue; (f) kidney tissue; (e, f) examples of the method validation (dots represent manually-assigned labels, green: labels classified as true positive, red: false negatives, yellow: false positives).

Figure 2. One dimensional grayscale image function I(x) with one dark (red) and one bright (green) object.

Initially, the minimum-model approach uses intensities to define objects (both hills and valleys in the intensity landscape can be objects).

Figure 3

(a) H&E stain of breast tissue, (b) local minima, (c) local maxima and (d) maximum local gradients of the horizontal image function I(x) marked with black pixels.

Figure 4

(a) 400×400 pixel H&E stained breast cancer tissue, (b) non-overlapping segmentation generated from full contour search (29815 contours in primary segmentation), (c) optimized contours, (d) results after concave object separation and (e) the final segmentation after removing all non-nuclei objects yielding 119 nuclei. The overall segmentation process took 0.39 seconds on a standard PC.

Figure 5. Image object (all pixels) with non-compact pixels on the object border (white pixels).

Numbers represent the distance of the corresponding pixel to the nearest background (non-object) pixel with the Manhattan metric. A distance value d is given for the testing of compactness (3 in this example). Removing all pixels p_i (with d_i the distance value of p_i and d_i < d) that are connected to a pixel p_j (with d_j = d) over more than d − d_i edges results in a compact object (gray pixels).

Figure 6. Cluster composed of two cell nuclei.

(a) Contour pixels (black), vertices of the convex hull (green), pixels of concave contour segments A₁B₁ and A₂B₂ (blue) and start and end point of potential cutting line C₁C₂ (red). (b) Cell nuclei separation result.