An automated neural-fuzzy approach to malignant tumor localization in 2D ultrasonic images of the prostate

Abstract

In this paper, a new neural-fuzzy approach is proposed for automated region segmentation in transrectal ultrasound images of the prostate. The goal of region segmentation is to identify suspicious regions in the prostate in order to provide decision support for the diagnosis of prostate cancer. The new automated region segmentation system uses expert knowledge as well as both textural and spatial features in the image to accomplish the segmentation. The textural information is extracted by two recurrent random pulsed neural networks trained by two sets of data (a suspicious tissues' data set and a normal tissues' data set). Spatial information is captured by the atlas-based reference approach and is represented as fuzzy membership functions. The textural and spatial features are synthesized by a fuzzy inference system, which provides a binary classification of the region to be evaluated.

Fig. 1

The segmentation problem: original TRUS image of the prostate (left panel) and manual segmentation of a malignant region as performed by a radiologist (right panel).

Fig. 2

System architecture.

Fig. 3

RNN neuron grid, the pixel is mapped to its first and second degree neighbors (N, NE, E, SE, S, SW, W, and NW).

Fig. 4

Synthetic image consisting of four natural textures.

Fig. 5

Results of the RNN a after filtering out the unconnected blocks and b after the results from a were subjected to median filtering.

Fig. 6

Effect of the size of the input block: a results from 4 × 4 input blocks and b results from 5 × 5 input blocks.

Fig. 7

Desired output with respect to a the distribution of the differences in RNN output (on y-axis, 1 signifies suspicious and 0 signifies normal), b the degree to which membership distribution is slotted into uneven-sized bins, and c the rescaled degree of membership distribution.

Fig. 8

Degree of membership (malignant) of input blocks in the ahorizontal direction and bvertical direction.

Fig. 9

Output membership functions μD1 and μD2.

Fig. 10

ROC analysis of segmentation results.

Fig. 11

Segmentation results produced by the proposed algorithm for predefined thresholds.

Fig. 12

Suspicious regions for sample images A, B, and C, identified by the Gabor-based algorithm.