Detecting distorted and benign blood cells using the Hough transform based on neural networks and decision trees - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 391

(7)
Specificity refers to the rate of correctly classified negatives, and it is equal to the ratio of TN
to the sum of TN and FP [ 9 ].
(8)
7 The experimental results
As mentioned above, the experimental results are displayed in two parts. The first part is con-
cerned with the detection and segmentation of RBCs in order to distinguish between benign
and distorted cells (caused by sickle-cell anemia), with the cells counted automatically. The
second part involves checking the previous detection process exported from part one by pre-
dicting and categorizing the segmentation results using the two most famous classiication
models in data analysis: the NN and the C&R tree [ 6 , 21 ] .
In part one, the detection process begins with importing and reading the microscopic
colored image of RBCs. All cells (benign or distorted) are then detected using CHT, a water-
shed process, and morphological techniques for enhancing the detection process. In the same
way, CHT is applied to the conditions of cell polarity to determine all dark and bright cells
according to their intensity. A two-stage technique subsequently computes the accumulator
array of the CHT. The sensitivity of this accumulator array for the proposed algorithm is 0.97
for brightness and 0.90 for darkness, and the edge gradient threshold is set at 0.2 to detect few-
er cells with weak edges. Actually, these conditions help the CHT to detect most of the benign
RBCs (near to the circle in shape), those positioned singularly or overlapping, and even those
attached to other distorted cells. Figure 6 shows the original image of RBCs in (a), and the pro-
posed algorithm is illustrated in detail from (b) to (f).