Comprehensive Morphometric Analysis of Apple Fruits and Weighted Class Assignation using Machine Learning

Abstract

Abstract Fruit morphology description for variety registration or evaluation is mostly based on human visual inspection. However, the development of an objective and efficient method for evaluating apple fruit shape would be of significant value. Furthermore, if this method can provide a comprehensive assessment of the multiple attributes encompassed by the term “shape”, it would have great potential for genomic studies. Here, we investigated the potential of a shape analyzer software originally developed to study tomato fruits (Tomato Analyzer) for the morphometric description of apple fruits. We conducted an analysis of 12,920 images of apple sections from 364 genotypes, collected across three harvest seasons. Also, we assigned the images into classes by visual inspection. The software detected the contour of the fruits in most of the images, but with some degree of imprecision, particularly in the stalk and calyx regions. After manual correction of the contours, we obtained 15 measurements of shape and size attributes. In general, size traits had higher heritability (H2) than shape traits (0.72 vs 0.45 in average, respectively). A Random Forest model was used to identify the most important variables determining fruit shape. The fruit shape index external I (FSII) outstood in importance, followed by the fruit shape triangle (FST), the distal angle Macro (DAMa), the eccentricity (ECC), and the proximal angle macro (PAMa). Incorporating these parameters into fruit description guides could provide more precise descriptions of apple cultivars. Additionally, this data will be useful to investigate the potential genetic control of these traits through genomic studies.