Faculty: Engineering
Department: Electronic And Computer Engineering


Chijindu, V. C.
Inyiama, H. C.


This work designed and developed an algorithm for an expert system based prostate disease diagnosis using image segmentation. The algorithm used segmentation of samples of trans-rectal ultrasound images of the prostate gland to perform image analysis and thus produce a diagnosis result. The algorithm analysed the images in two stages to arrive at the result of diagnosis. First, the boundary of the prostate gland was detected and secondly the regions in the three zones of the gland were scanned for hyperechoic pixels in clusters which indicate sections of the gland affected by cancer. Enhanced region growing segmentation algorithm was employed for the two stages. Radial/axial scanning of gland pixels were carried out from a common centre automatically selected. Expert knowledge was elicited and implemented in the segmentation algorithm; hence the entire system was modelled as an expert system. Structured system analysis and design methodology was adopted for designing and developing the various modules that realised the objectives of this work. MATLAB programming tool was used to code the developed algorithms. Samples of TRUS 2D -images of the prostate for patients with raised PSA values (>10 ng/ml) used in a previous work by Award (2007) were used for testing the algorithm. The test results showed that the algorithm could detect the prostate boundary, detect the zones of the gland that exhibit image properties for cancer cells where they occurred in the different samples. The processing time for the algorithm was approximately 40s. The algorithm was validated by using area-based metrics involving expert’s results. Accuracy and sensitivity values were computed for the two stages. An average accuracy of 88.55% and sensitivity of 71.65% were recorded for segmentation of prostate gland while accuracy of 84.17% and sensitivity of 78.49% were recorded for detecting the hyperechoic (suspected cancer cells) regions of the prostate gland. The results compare well with an earlier work using parametric deformable model technique on the same life image samples, which recorded average accuracy of 91.03% and sensitivity of 95.60% for segmentation and average accuracy of 84.16% and sensitivity of 66.56% for detecting the hyperechoic regions of the prostate gland. With such high values of accuracy and sensitivity recorded it was concluded that enhanced region growing technique can be used to successfully detect segments and sections of prostate gland with cancer cells from TRUS 2D-images