TURBOCOW™ completes the development of mastitis diagnostics system with promising results
Three years ago, SIA “Vet Health Spektrum” (TURBOCOW™) started the implementation of the project “Industrial Research of the Artificial Intelligence Based System”, the activities of which are nearing completion. The Faculty of Veterinary Medicine of the Latvia University of Agriculture, Scientific Laboratory of Biotechnology of the Latvia University of Agriculture, The Institute of Solid State Physics of Latvian University, the intensive dairy farm Agrokaķinieki, biofarm Bikstu Brīvzemnieki, the full cycle dairy farm Zilūži, and Latvian Cooperation Council of Agricultural Organizations (LOSP) have been involved as partners. The aim of the project was to develop, test and validate a newly developed artificial intelligence-based diagnostic system for mastitis in order to increase productivity efficiency in the dairy sector and to promote the development of the sector as a whole.
It was very important is to find out the method for quick detection of udder health with the use of low electromagnetic waves biosensor system (BSS). The main task was to compare BSS readings at the extreme states of the significance of mastitis pathogens, namely, `no mastitis pathogen` or `significant growth of mastitis pathogen”. The present study clearly shows that both conditions have a decisive effect on the direction of mastitis development. A mammary gland free of mastitis agents might show a drop in SCC from several million to 100 thousand cells per mL of milk in one month. Also, BSS’s ability to predict future changes in SCC was estimated (κ = 0.32), but it does not reach the moderate coincidence level (κ> 0.40). However, the SCC level can be successfully predicted with high accuracy (κ = 0.89) if, udder quarter milk samples are tested for the presence of a mastitis agent and the result is serially interpreted with the results of the California mastitis test and clinical investigation. The accuracy accuracy of 93% has been achieved during the CNN model estimation on the test dataset via the cross-validation method.