Vetnews | Oktober 2024 14 « BACK TO CONTENTS Article Statistical framework of heterogeneity Cochran’s Q test is the traditional test for heterogeneity in metaanalyses. This allows for the idealization of the study size within a desired level of precision. Cochran formula: Z2pq n0 = e 2 In the equation, e is the level of precision; p is the proportion of the population; q = 1 − p; and the value of Z is given in Z-table [7, 8]. To set a target population, we need to determine how many AI articles must be reviewed to have maximum variability (randomized trial). Using a 95% confidence limit and p = 0.05, we found that a random sample of 385 articles in our target population should be sufficient to give the desired confidence levels. Quality assessment (risk of bias) The scientific quality of each article was assessed according to the Cochrane guidelines [7]. The result of the assessment is noted in Table 2. A standardized form containing nine criteria was developed to assess each paper’s overall quality and risk of bias. The quality rating scale contained nine items and was rated by two independent reviewers on a binomial scale and was summed to give an overall indication of quality [7]. Results Overview The initial electronic search identified 883 references. After screening and selection, 812 studies were included in the study and 71 were excluded from the study. Based on the analysis noted above, a random sample of 385 of the 812 identified articles should be sufficient to give the study a 95% confidence level. There were four major areas that encompassed AI implications in the veterinary field. The first area was diagnostic studies. The second area was papers involving education, animal production, and epidemiology with an equal number of papers on each topic. The third area included animal health and welfare, pathology, microbiology, and duplicated citations with an approximately equal number of papers on each of these topics. The remaining area consisted of papers in all other subcategories that did not reach the threshold of 30 papers each. A detailed scheme of this breakdown of topics and areas is noted in Figure 2. Systematic search In fact, 150 articles were eventually excluded following further retrospective analysis. A total of 79 citations were retrieved based on their full-text analysis. Descriptive analysis Figure 1 summarizes the included characteristics of this study. The 812 works were published up to March 22, 2023, and these studies were conducted worldwide. A total of 192 studies were categorized as diagnostic, 93 as education, 91 as animal production, 86 as epidemiology, 63 as animal health and welfare, 55 as pathology, and 33 as microbiology. As noted above, the remaining studies did not qualify for a separate category. These remaining diverse studies included toxicology, 14; pharmacology, 13; oncology,21; haematology, 17; anatomy, 3; nutrition, 18; anaesthesia, 15; statistics, 17; environment and ecology, 25; biochemistry, 16; histology, 23; and embryology, 17. In addition, 79 citations were added to the included data after a full-text review of the search results found that AI was also referred to by the terms ML, CNN, and DL. Machine learning, CNN, and DL were referred to 36.7%, 16.45%, and 46.83% of the time, respectively. Characteristics of the included studies During the course of reviewing the selected articles, their categorization could change. For instance, studies related to hematology covered a wide range of diseases and blood abnormalities such as blood cancer, blood analysis, and hereditary or genetic diseases. The classification of these topics changed as we removed studies related to cancer from hematology and added them to oncology after all co-authors agreed to such a move. All articles were given a category such as epidemiology or diagnosis and a subcategory such as blood cancer, risk factors, infectious diseases, or metabolism. Subsequently, a final classification was made based on unifying subfields in the same category. Category selection The documents were classified as peer-reviewed studies, journal/ book papers, or conference reports during the collection and review process. Particular attention was paid to peer-reviewed studies (65%), as they detailed their research process as opposed to conference reports (10%), which focused more on results. Table 3 categorizes all of the collected works. A meta-analysis of these studies was not possible due to the scarcity and lack of raw data required to determine accepted accuracy measures, but also due Table-1: Standardized conceptions for studies categorization. Category number Category nomination Category description 1 Microbiology All articles related to bacteriology, virology, and mycology. 2 Diagnostic Imaging, MRI, CT scanning, and different testing. 3 Epidemiology Infectious diseases (spreading, distribution, and surveillance), risk factors, monitoring systems, prevention, and forecasting. 4 Animal Health and welfare Treatment, drugs, surgeries, ethics, and animal well-being. 5 Education e-learning, administrative support, and teaching process. 6 (Digital) pathology Different pathogenic agents, pathogenicity, histopathology, and physiopathology. 7 Animal reproduction Farming management breeding of all species. 8–19 Divers Includes toxicology, pharmacology, oncology, hematology, anatomy, nutrition, anesthesia, statistics, biochemistry, histology, embryology, and ecology. MRI=Magnetic resonance imaging, CT=Computed tomography
RkJQdWJsaXNoZXIy OTc5MDU=