br R Nurfauzi et al
8 R. Nurfauzi et al. / Journal of King Saud University – Computer and Information Sciences xxx (xxxx) xxx
Fig. 21. The model of point representation in small nodule by using ABM-corner and ABM-local minima-maxima (Nurfauzi et al., 2017).
segmentation nodule areas (Nurfauzi et al., 2017) are necessary to overcome. Corner feature represents the curvature in lung bound-ary. Instead of using all pixels in the lung boundary, the computa-tional time can be significantly reduced by using the feature of the nodule basin location only. Performance comparison of the pro-posed method with other methods is summarized in Table 1.
In Table 1, there are four evaluation components, namely under-segmentation of nodule area (US), over-segmentation of nodule area (OS), false positive of nodule basin and computational time. The higher the US value, the wider the area of nodule that is not covered by the system. The higher the OS value, the wider the error of overlapping area. False positive of nodule basin indicates the incorrectly detection of nodule basin by the system. This affects in the addition of the number of new suspected nodules in lung boundary. The increase in the number of detected nodule basin results in the increase in the number of new suspected nodules.
The simple and the common method to cover nodule basin is morphological operation. Based on the simulation result as shown
in Table 1, morphological operation produces suitable result with value of US, OS and computational time are 15.2%, 114% and 0.52 s/slice, respectively. However, lots of FPs still occur as shown in Fig. 20(b) indicated by cyan arrow. To reduce the number of FP, ABM conventional is adopted. In this Cell Counting Kit-8 (CCK-8) cck8 cck-8 study, the number of FP and the US value of ABM conventional are better than that of morpho-logical operation with 17.5 and 10.6%, respectively. Nevertheless, ABM conventional still produces high OS value and needs more time to execute the data.
To reduce the computational time of ABM conventional, loca-tion represented the nodule basin should be found. Local minima and maxima features are used. According to Table 1, ABM min-max successfully reduces computational time and OS value. How-ever, this method still has some limitations. ABM min-max only represents the curve in x and y directions. Consequently, antidiuretic hormone (ADH) method cannot detect the nodule basins that are not curved in the x and y directions. Nevertheless, this situation often occurs in small nodule basin size. Hence, many small basins cannot be detected by this method. The created model in Fig. 21 shows that the local minima-maxima features which are pointed as red squares are not sensitive to represent the nodule basin. Hence, the ABM min-max result produces high US value of 16.2% and less FP of 8.2. The lowest FP occurs because the feature is not able to detect small curvature. Mostly, the small curvature in lung boundary represents noises.
To detect small nodule basin, the features that are able to ana-lyze in all directions are essential. One of these features is corner feature. Corner feature can be effectively extracted to represent nodule basin. The points representing corner feature are depicted with blue squares in Fig. 21. The proposed method is suitable to detect the small nodule basin and to represent the nodule basin properly. With this ability, the proposed method achieves the bet-ter performance than ABM min-max and morphological operation indicated by value of US. Moreover, the proposed method also
Fig. 22. The comparison of small juxta-pleural covering performance. (a, f, k) are the original CT images; (b, g, l) are adaptive border correction results of morphological operation; (c, h, m) are results of ABM conventional; (d, i, n) are results of ABM corner and (e, j, o) are results of ABM-local minima-maxima.
Please cite this article as: R. Nurfauzi, H. A. Nugroho, I. Ardiyanto et al., Autocorrection of lung boundary on 3D CT lung cancer imagesq, Journal of King Saud University – Computer and Information Sciences, https://doi.org/10.1016/j.jksuci.2019.02.009
R. Nurfauzi et al. / Journal of King Saud University – Computer and Information Sciences xxx (xxxx) xxx 9
achieves the fastest computational time compared to other methods.
Theoretically, a method that is able to reduce US and OS values can detect nodule basin more accurately. To implement the method into an embedded system also requires faster computa-tional time. The proposed method successfully elaborates these abilities. Small number of FP does not indicate that a method has good ability in detecting nodule basin as illustrated in Fig. 22. In this case, small number of FP indicates that the ABM min-max is not sensitive to small curvature.