Explainable Deep Learning Reproduces a 'Professional Eye' on the Diagnosis of Internal Disorders in Persimmon Fruit.

Recent rapid progress in deep neural network techniques has allowed recognition and classification of various objects, often exceeding the performance of the human eye. In plant biology and crop sciences, some deep neural network frameworks have been applied mainly for effective and rapid phenotyping. In this study, beyond simple optimizations of phenotyping, we propose an application of deep neural networks to make an image-based internal disorder diagnosis that is hard even for experts, and to visualize the reasons behind each diagnosis to provide biological interpretations. Here, we exemplified classification of calyx-end cracking in persimmon fruit by using five convolutional neural network models with various layer structures and examined potential analytical options involved in the diagnostic qualities. With 3,173 visible RGB images from the fruit apex side, the neural networks successfully made the binary classification of each degree of disorder, with up to 90% accuracy. Furthermore, feature visualizations, such as Grad-CAM and LRP, visualize the regions of the image that contribute to the diagnosis. They suggest that specific patterns of color unevenness, such as in the fruit peripheral area, can be indexes of calyx-end cracking. These results not only provided novel insights into indexes of fruit internal disorders but also proposed the potential applicability of deep neural networks in plant biology.

[Presumption of the energy-spectrum of high-energy electron beam based on the beta-distribution model].

The energy spectra of high-energy electron beams used in radiotherapy are the most important data for evaluating absorbed doses and/or dose distributions in the body of a patient. However, it is impossible to measure the actual spectra of a high-energy electron beam. In this study, we suggest a method to presume the spectra of high-energy electron beams by use of the beta distribution model. The procedure of this method is as follows: (1) the spectrum of the high-energy electron beam was assumed to have a maximum energy Emax, and α, ß parameters of the beta probability density function. (2) The percentage depth dose (PDD) based on the assumed spectrum was calculated by a Monte Carlo simulation. (3) The best matching energy spectrum was searched in comparison with the experimental PDD curves. Finally, the optimal energy spectrum of the electron beam was estimated after reiterating the process from (1) to (3). With our method, the measured PDD curves were optimally simulated following the experimental data. It appeared that the assumed spectra approximated well to the actual spectra. However, the error between the assumed and experimental data was observed in the region under the incident surface. We believe this was due to the influence of low-energy electrons scattered at installed collimators, etc. In order to simulate PDDs in this region accurately, a further correction process is required for a spectrum based on the beta distribution model.