Detecting internal browning in apple tissue as determined by a single CT slice in intact fruit

Internal browning (IB) in apples is associated with small voids that appear dark in X-ray computed tomography (CT) scans. Previous reports investigating IB in apples have used high resolution CT. However, for high-throughput in-line sorting and grading, detection of IB using CT, the image resolution and number of CT slices analysed must be limited. This work aimed to determine if IB can be accurately detected using a single CT slice with an image resolution that is not high enough to detect individual cells and pores in whole ‘Kanzi’ and ‘Braeburn’ fruit. Whole ‘Kanzi’ fruit (n = 120), previously stored under controlled atmosphere conditions (4 °C, 1 kPa O2, 1.5 kPa CO2) for eight months, were scanned with CT at a voxel size of approximately 130 µm. Whole ‘Braeburn’ fruit (n = 79) were stored under browning inducing conditions (0.5 °C, 1.5 kPa O2, 5 kPa CO2) for 2 and 3 months before scanning. For both cultivars, fruit (‘Kanzi’ = 20, ‘Braeburn’ = 38) that were stored under regular air conditions for one month were included as a non IB control. Following scanning, fruit were visually scored on browning intensity and type. An image-processing procedure based on grey-level threshold values was developed to determine the area percentage of detectable voids in CT slices. For both cultivars, the void percentage (Vp) in CT slices increased as the browning intensity in apple flesh increased. The largest variation in Vp among browning intensity classes was within the first 20 mm from the fruit stem end for ‘Kanzi’ fruit and between 30 and 40 mm for ‘Braeburn’ fruit. These results were due to the high incidence of radial flesh browning (RFB) within ‘Kanzi’ samples, which occurred concurrently with core browning at equal intensity. For ‘Braeburn’ fruit, RFB occurred in fruit with severe core browning. Logistic regression modelling using the Vp of CT slices at 30 and 35 mm of the fruit stem end showed better performance [area under the receiver operating characteristic curve (AUC) > 0.9] for classifying affected ‘Kanzi’ and ‘Braeburn’ fruit than models using the whole fruit average. These results demonstrate that the entire fruit does not need to be analysed for accurate classification. Using only one slice per fruit for browning detection would greatly reduce image analysis time, which is imperative for in-line IB detection using CT.