Archives

  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • 2021-03
  • br To verify the e ectiveness of the

    2020-08-12


    To verify the effectiveness of the SPA-PAL2v algorithmic structure, even with the use of the total samples to create the paraconsistent patterns, two methods were adopted to analyze the histopathological
    Fig. 5. (A) Raman spectra of sample (no. 162) randomly selected from among 223 total samples; (B) similarity points between the study sample and the para-consistent pattern for normal Apicidin tissue; (C) similarity points between the study sample and paraconsistent pattern generated for the BCC and SCC group; and (D) similarity points between the study sample and the paraconsistent pattern for AK skin tissue.
    Table 2
    Test results of arrangement (NO × [BCC + SCC] × AK) for Mode 1 (without cross-validation).
    Arrangement of Agreement Disagreement Correct Histopathological Groups
    Classification
    Global Accuracy 76%
    arrangements of the group. In the first method, Mode 1 (without cross-validation), the analysis was performed using the total samples to create the paraconsistent patterns. It can be observed from Table 2 that the lesions of non-melanoma cancer (BCC + SCC) had the highest dis-agreement value (38). This may be due to the extraction process of the Raman spectrum; the laser might have focused on the surrounding re-gions of the main lesion. Although this disagreement does not in-validate the information, it distorts another clinical fact, as it suggests that the SCC lesions are in the initial phase, and thus, they can be confused with tumor lesions of type AK. Due to the above phenomenon, the (BCC + SCC) arrangement corresponded to a classification accuracy of 72%.
    In the second method, Mode 2 (with cross-validation), sample subsets were selected separately from those that were used to create paraconsistent patterns. As presented in Table 3, the subset 2 presented the worst accuracy (partial accuracy 3 = 65.9%), with a difference of 18.2% in relation to the best value obtained (partial accuracy 1 = 84.1%). Subset 3 had the second-worst accuracy (partial accuracy 3 = 70.3%), with a difference of 13.8% in relation to the best value obtained (partial accuracy 1 = 84.1%). These differences suggest that in future works, the paraconsistent patterns of the samples belonging to subsets 2 and 3 in the arrangement should be carefully studied. With the investigation of the profile differences and the amplitudes of the samples belonging to these subsets, new information can be obtained for a complete optimization of the algorithms, and thus, a better re-sponse of the SPA-PAL2v will be obtained. Both methods present a very similar global accuracy, which indicates that the standards obtained through the algorithms PAL2v do not result in overtraining effects. 
    Table 3
    Arrangement of Agreement Disagreement Correct Histopathological Groups
    Classification
    Accuracy
    The result of the analysis using the PCA/DA method is presented in Table 4. The SPA-LPA2v was verified to be was superior in the two modes (Mode 1 –without cross-validation and Mode 2 –with cross-va-lidation).
    The percentage accuracy results obtained by SPA-PAL2v were compared with the results presented in reference [19], in which five statistical techniques were used with a database of Raman samples of skin tissues obtained in vivo but grouped in the same format as in this work. The statistical techniques used in [19] include PCA/DA (with
    Table 4
    Percentage of accuracy in the arrangement obtained by the PCA/DA statistical methods and SPA-PAL2v in Mode 1 and Mode 2.
    Global accuray
    principal components analysis Paraconsistent Logic (PAL2v) Mode 1- without Paraconsistent Logic (PAL2v) Mode 2- with cross
    discriminant analysis cross validation validation Arrangement Type PCA/DA SPA-PAL2v SPA-PAL2v NO × (BCC + SCC) × AK 74 % 76 % 75.78 %
    discrimination done by Euclidean, quadratic and Mahalonobis dis-tances) and partial least squares with discriminant analysis (PSL/DA). For the arrangement (NO × [BCC + SCC] × AK), the results obtained in [19] by the statistical methods are as follows: PCA/DA-Euclidean distance resulted in 74.7%; PCA/DA quadratic distance, 73.7%; PCA/ DA Mahalanobis distance, 74.7%; and PLS/DA, 82.7%. The results do not correspond to the same database of this study, but they can be considered valid since they address the same lesions and arrangement type. The percentage accuracy values obtained in this study with PAL2V (Table 4) surpassed the results of three PCA statistical techniques pre-sented in [19].
    A first analysis of the dataset revealed prominent spectral differ-ences existed among the non-melanoma tumors and premalignant and normal tissues (plots not shown). As previously demonstrated by sev-eral studies, these differences are related to differences in the relative concentrations of proteins and lipids, depending on the tissue type (non-tumor or non-melanoma lesion); thus, spectral analysis by PL to identify these spectral features is difficult.