75 AI-Enhanced Hematocrit Analysis: Revolutionizing rapid anemia detection in small ruminants

Abstract

Abstract In recent times, there has been a growing focus on the well-being and healthcare of small ruminants, particularly in relation to the issue of anemia due to infection with blood-feeding gastrointestinal nematodes, such as Haemonchus contortus. The objective of this study was to assess the hematocrit levels in blood samples obtained from small ruminants, specifically goats. Additionally, the study is an attempt to design and create a quick sensor for identifying anemia, which could be conveniently used on farms. A total of 75 mature, intact male Spanish goats were subjected to hematocrit analysis to ascertain their spectrum of hematocrit values and association with anemic conditions. Simultaneously, a unique sensor with user-friendly features was developed to promptly provide farmers with feedback on the anemic status of animals using AI-based machine learning algorithms, enabling timely intervention. The sensor utilized a semi-invasive approach with minimal blood sample requirement. 30 µL of blood was dropped on glycerol soaked Whatman filter paper No. 1, and images were taken at 90 sec, 150 sec, and 270 sec to capture the blood pattern on filter paper soaked in glycerol using a cell phone. These images were correlated with actual hematocrit value obtained from conventional hematocrit analysis. A RMSprop with adjusted learning rate convolution neural network (RMSprop-CNN) based images classification model was developed for the classification of different patterns at different levels of hematocrit. The model was trained on a total of 1,000 images with a training and testing split of 80:20 ratio. For the optimization process, the Adam optimizer with a learning rate of 0.001 is employed. The model is compiled with a categorical cross-entropy loss function, aiming to improve its accuracy metric over training iterations. The initial studies demonstrated a detection accuracy level of 70.31 % at 10 epochs in identifying different hematocrit levels (Level 1; healthy, to Level 5; severely anemic) but improved significantly up to 94.89 % by 100 epochs, resulting in a notable reduction in the time and level of knowledge previously necessary for conducting such evaluations. The present study not only provides valuable insights into the hematological characteristics of small ruminants, but also establishes a foundation for the development of a straightforward and efficient method for detecting anemia. Consequently, this advancement has the potential to enhance the overall care and well-being of animals within agricultural environments.