We compared two techniques of machine learning for the identification of cows that will be good producers of milk based on their genome-wide information. Data from a genome-wide genotyping panel, consisting of 164312 single nucleotide polymorphism markers (SNPs), within the 29 autosomal chromosomes, from 1092 Holstein cow samples were used for this study. Sample cows were divided as high-milk producers and lowmilk producers based on their estimated breeding value of the 305 day average milk yield. Seven data sets were generated that grouped chromosomes with the highest number of SNPs related to milk production for prediction. Decision trees and artificial neural network algorithms were trained and tested, and the performance of prediction was computed. The mean prediction accuracy obtained with the decision tree algorithm was 92.44%, with a maximum of 94.5%, while the mean prediction accuracy obtained with the artificial neural network algorithm was 82.19%, with a maximum of 87.3%. Also, the decision tree algorithm permitted the identification ©FUNPEC-RP www.funpecrp.com.br Genetics and Molecular Research 18 (4): gmr18407 E. Rodríguez et al. 2 of the most dominant single nucleotide polymorphism for prediction, which is situated within a milk-related quantitative trait locus in chromosome 14. Finally, our results add new evidence to support that machine learning algorithms may be used for managing genomewide SNP markers, for implementing classification and prediction tools in the cattle industry.

No abstract

Deep learning methods have become the state of the art for undersampled MR reconstruction. Particularly for cases where it is infeasible or impossible for ground truth, fully sampled data to be acquired, self-supervised machine learning methods for reconstruction are becoming increasingly used. However potential issues in the validation of such methods, as well as their generalizability, remain underexplored. In this paper, we investigate important aspects of the validation of self-supervised algorithms for reconstruction of undersampled MR images: quantitative evaluation of prospective reconstructions, potential differences between prospective and retrospective reconstructions, suitability of commonly used quantitative metrics, and generalizability. Two self-supervised algorithms based on self-supervised denoising and the deep image prior were investigated. These methods are compared to a least squares fitting and a compressed sensing reconstruction using in-vivo and phantom data. Their generalizability was tested with prospectively under-sampled data from experimental conditions different to the training. We show that prospective reconstructions can exhibit significant distortion relative to retrospective reconstructions/ground truth. Furthermore, pixel-wise quantitative metrics may not capture differences in perceptual quality accurately, in contrast to a perceptual metric. In addition, all methods showed potential for generalization; however, generalizability is more affected by changes in anatomy/contrast than other changes. We further showed that no-reference image metrics correspond well with human rating of image quality for studying generalizability. Finally, we showed that a well-tuned compressed sensing reconstruction and learned denoising perform similarly on all data. The datasets acquired for this paper will be made available online; see <a href='https://www.melba-journal.org/papers/2022:022.html'>https://www.melba-journal.org/papers/2022:022.html</a> for details.