A New Collaborative Classification Process for Microcalcification Detection Based on Graphs and Knowledge Propagation

Abstract

In this paper, we propose a new collaborative process that aims to detect macrocalcifications from mammographic images while minimizing false negative detections. This process is made up of three main phases: suspicious area detection, candidate object identification, and collaborative classification. The main concept is to operate on the entire image divided into homogenous regions called superpixels which are used to identify both suspicious areas and candidate objects. The collaborative classification phase consists in making the initial results of different microcalcification detectors collaborate in order to produce a new common decision and reduce their initial disagreements. The detectors share the information about their detected objects and associated labels in order to refine their initial decisions based on those of the other collaborators. This refinement consists of iteratively updating the candidate object labels of each detector following local and contextual analyses based on prior knowledge about the links between super pixels and macrocalcifications. This process iteratively reduces the disagreement between different detectors and estimates local reliability terms for each super pixel. The final result is obtained by a conjunctive combination of the new detector decisions reached by the collaborative process. The proposed approach is evaluated on the publicly available INBreast dataset. Experimental results show the benefits gained in terms of improving microcalcification detection performances compared to existing detectors as well as ordinary fusion operators.

Keywords: Collaborative classification; Graph knowledge propagation; Mammography; Microcalcification detection.

Fig. 1

Global flowchart of the proposed approach

Fig. 2

Samples of SP1- and SP2-generated superpixels (blue contours) with superimposition of MC contours (red contours)

Fig. 3

Conversions of the thematic and suspicion maps

Fig. 4

Different possible scenarios when identifying candidate objects from two thematic maps (red and green contour regions)

Fig. 5

Samples of identifying candidate objects from three different suspicious areas

Fig. 6

Sample of an initial classification of the candidate objects represented by their positions in the image (gravity center): $C_{\mathit{AC}}$, $C_{\mathit{PC}}$ and $C_U$ objects are respectively colored by red, green, and blue colors

Fig. 7

Global flowchart of the collaborative classification process

Fig. 8

Illustration of a sample graph generated from a set of candidate objects. The objects that belong to the uncertainty class (blue circled objects) are not considered as nodes for the graph

Fig. 9

Sample of a group of MCs present in the same SP1 superpixel

Fig. 10

A sample of an intra-superpixel refinement

Fig. 11

A sample of an inter-superpixel refinement

Fig. 12

Illustration of the possible evaluation situations

Fig. 13

Percentages of the TP rates resulted obtained by the application of the proposed approach

Fig. 14

Boxplots of the obtained true positive rates from the intersection operator applied to the initial detector results and after the proposed collaborative process