A deep learning algorithm predicts milk yield and production stage of dairy cows utilizing ultrasound echotexture analysis of the mammary gland

For decades now, ultrasonography is being utilized in dairy animals' udder health care. Recent studies have employed echotexture analysis for the examination of the mammary parenchyma and the investigation of pathological conditions. The present study aimed to investigate whether a deep neural network could predict the cows' daily milk yield and production stage, utilizing features from echotexture analysis of mammary gland sonograms. A dataset of 3072 B-mode sonograms was created, after the ultrasonographic examination of Holstein dairy cows at successive stages of their production cycle from 2019 to 2021. Echotexture analysis was applied to these sonograms providing a dataset of features. This feature set was utilized as input for the construction of 5 models, based on the same deep neural network architecture. The first model aimed to binary classify the daily milk yield of dairy cows between High Lactation and No Lactation. The other four models aimed to binary classify the production stage of the cows between four pairs of stages: Dry Period - Peak Lactation, Late Lactation - Peak Lactation, Late Lactation - Fresh Cow and Peri-Partum - Peak Lactation. The High Lactation - No Lactation model achieved high performance, with the best execution reaching 91.3% Accuracy and 90.91% F1-score. Among the four production stage classification models, the Dry Period - Peak Lactation achieved the highest Accuracy and F1- score. The Late Lactation - Peak Lactation, Late Lactation - Fresh Cow and Peri-Partum - Peak Lactation models also presented promising results. These results indicate that a deep neural network can utilize features from mammary sonograms' echotexture analysis to predict a cow's daily milk yield and production stage. A larger dataset could lead to better performance of the models. Further progress could make the qualitative assessment of the mammary parenchyma possible and aid on-site diagnoses of pathological conditions, in the direction of smart dairy farming.

Automated summary

This study utilized echotexture analysis of mammary sonograms and deep neural network models to successfully predict the daily milk yield and production stage of dairy cows. The model for classifying High Lactation and No Lactation achieved the best performance. These results suggest the potential of deep neural networks in smart dairy farming and the qualitative assessment of mammary parenchyma.