Diagrammatic scales for the estimation of black node disease severity in common bean

: The objective of this study was to design two diagrammatic scales for the evaluation of black node disease (Stagonosporopsis hortensis and Boeremia spp.) in common bean (Phaseolus vulgaris L.). The developed scales corresponded to logarithmic intervals (SADL) of seven levels and equal intervals (SADE) of nine levels. These scales were compared to the scale developed by the International Center for Tropical Agriculture (CIAT), which corresponds to a descriptive scale (NoSAD). In the scale validation the accuracy, precision, and concordance of repeatability were determined using Lin’s concordance correlation coefficient, whereas inter-rater reproducibility was determined by the overall concordance correlation coefficient (OCCC). It was observed that with the use of the proposed scales, reproducibility and repeatability were significantly improved for inexperienced raters, and reproducibility was improved in the case of experienced raters. Thus, the designed standard area diagrams with equal and logarithmic intervals are a useful tool for estimating severity under field and experimental conditions as part of the study of this patho-system.


Miranda et al.
Although, some scales to evaluated black node disease in common bean have been developed, these scales have not been validated (STONEHOUSE, 1994;CASTAÑO, 2002).
The International Center for Tropical Agriculture developed an ordinal and descriptive scale, with nine severity levels (SCHOONHOVEN & PASTOR-CORRALES, 1987), which is called NoSAD in the present study. Although, this scale does not have a description for even levels neither a standard area diagram, is currently widely being used in the evaluation of bean germplasm.
Thus, the objective of this study was to develop and evaluate two quantitative diagrammatic scales with photographic support from leaves of commercial crop plants. These scales were designed using both equal (SADE) and logarithmic (SADL) intervals. They were compared to the nondiagrammatic ordinal scale (NoSAD) that CIAT previously established in order to determine if there were significant differences in accuracy, precision, and reliability in the estimates obtained with the use of each scale.
Two hundred leaves of plants that showed different degrees of disease severity were collected in different commercial bean crops in Colombia. Each of these samples was photographed with a digital camera (Nikon D3400, 24.2 megapixels). Image analysis was performed individually using the ImageJ program (ImageJ, NIH, Bethesda, ML). The limits of the scales were established, considering into account the maximum disease severity reported in the plant material with the entire leaf area. The intervals of the scales were defined by two methodologies: logarithmically as established by HORSFALL & BARRATT (1945) and with intervals of equal amplitude as proposed by NITA et al. (2003), maintaining the nine levels of the standard system for the evaluation of bean germplasm determined by SCHOONHOVEN & PASTOR CORRALES (1987).
Scale validation was performed with five raters with experience and five raters without experience in disease evaluation, using the NoSAD scale and the SADL and SADE scales proposed in the present study. Forty randomly projected images covering different severity levels were evaluated. This exercise was performed three times to estimate its repeatability for each scale, with a minimum of seven days between each one.
Lin's concordance correlation coefficient (LCCC (pc)) was calculated to assess precision and accuracy. To establish if there were differences between the means of the LCCC statistics, a bootstrap of 10,000 repetitions was performed. The mean was calculated, and the equivalence test was performed, taking into account the 95% confidence interval (CI) from the difference between means. To evaluate the intra-rater reliability, Lin's coefficient of the estimates obtained by each rater was calculated. The mean reliability was determined, and it was established if there were significant differences between experienced and inexperienced raters when comparing the first and second, and the second and third repetitions. Regarding reproducibility of the estimates between raters, the overall concordance correlation coefficient (OCCC) was calculated for each repetition (CORREIA et al. 2017). The same procedure described for LCCC statistics was performed to determine if there were significant differences.
The lower and upper limits were established with values of 0.3% and 54%, respectively. GODOY et al. (2006) define the upper limit as the amount of disease in which leaf abscission occurs. In the reported bean scales, this value corresponds to 24% (GODOY et al. 1997) and 51% of leaf area affected (DÍAZ et al. 2001). Although, the maximum value of actual severity was 54%, it is possible to find higher values, however, their estimation is difficult since the detachment of one or two leaflets is frequent, and there is a presence of wrinkled or rolled leaves. Likewise, other pathogens can be reported in advanced stages of infection, so the actual estimation of leaf area affected by the causative agent of the black node disease is limited. Thus, the maximum severity value observed is consistent with that reported by DEL PONTE et al. (2017), who estimated that the maximum severity value was less than 60% in half of the scales when analyzing 105 SAD.
Thus, the values of the first seven categories proposed by HORSFALL & BARRATT (1945), were used in the elaboration of the scale with standard area diagrams of logarithmic intervals (SADL), in which the maximum limit corresponded to values greater than 50% (scale not shown). In the case of the scale with standard area diagrams of equal intervals (SADE), nine categories were established with an interval size of seven, and thus the maximum limit was for severities higher than 49% (Figure 1).
Experienced raters did not show any significant differences in accuracy, with the use of the three scales, since the bias factor correction (Cb) values were close to 0.8 (Table 1). The opposite results were observed for inexperienced raters since significant differences were obtained when comparing the means of NoSAD to the ones of SADL and SADE. These results indicated that inexperienced raters assigned severity values closer to the actual value when they had a reference image. Regarding precision, there were no significant differences in the estimates when comparing the correlation coefficient (r) of the three scales in the two groups of raters (Table 1).
For experienced raters to the location shift (u) parameter, significant differences were reported when comparing NoSAD (0.4158) and SADL (0.5161) scales to SADE (0.6622), which suggested that the overestimation increased with this scale. On the contrary, for inexperienced raters, NoSAD (0.8592) and SADL (0.4564) showed significant differences, but they were not observed when compared to SADE (0.6097). Thus, inexperienced raters overestimated the disease more with the use of NoSAD, followed by SADE (Table 1). These results would confirm what NUTTER et al. (2006) stated, who indicated that the scales based on Horsfall and Barrat were not necessarily invalid, but required further analyses to corroborate the severity estimated by raters when using intervals. In this sense, the validation of the scale is essential as it is a process that allows comparing the estimates of disease severity obtained by the use of the scale versus the actual value, establishing if its use improves accuracy, precision, and reliability.
When comparing the results of Lin's coefficient of experienced and inexperienced raters for the NoSAD scale, significant differences were reported between the means in four of the five statistics of this coefficient. Experienced raters using NoSAD had higher concordance values, which means greater accuracy and precision compared to inexperienced raters. However, the concordance of the estimates by inexperienced raters when using SADE and SADL did not show significant differences compared to the estimates of experienced raters. Based on these results, it can be considered that the use of diagrams improves estimates by inexperienced raters up to the point of performing similarly to an experienced rater, as observed in other studies (YADAV et al. 2012;GONZÁLEZ-DOMÍNGUEZ et al. 2014;LAGE et al. 2015). It should be noted that inexperienced raters did not receive any training to perform severity estimation. Regarding inter-rater reproducibility, both experienced and inexperienced raters increased reliability with the use of SADE or SADL, which indicated that there is a more considerable agreement of the estimates performed between raters with the use of diagrammatic scales compared to the use of NoSAD (Table 2).
Thus, the scales proposed in this project, provided an advantage over evaluation with NoSAD scales for people who are not familiar with the disease. They also improved inter-rater reliability among experienced raters, which is essential in case of experiments that are carried out in different locations where several people are in charge of data collection. Additionally, it is necessary to introduce severity diagrams in the range from 0 to 7% since they support decision-making processes regarding the selection of materials for the study of disease resistance. Thus, it would be necessary to establish the thresholds for a bean plant that is resistant, intermediate resistant, and susceptible to the black node disease. However, based on the results obtained in the present study with the scales of diagrams, SADE is proposed as an alternative tool to the CIAT scale since it enables the obtaining of a disease estimate in a more exact and precise way. Furthermore, it maintains the nine standard levels that can be useful for people who are interested in bean cultivation and black node disease research. Table 1 -Lin's concordance correlation coefficient (LCCC). Concordance of the estimates of black node disease severity on bean leaves carried out by experienced and inexperienced raters using the NoSAD, SADL, and SADE scales.