Vision Transformer-Based Blood Group Classification on Slide Images

Blood is an essential fluid in the human body, enabling the transport of oxygen and nutrients. In cases of accidents, persistent hematological illnesses, or surgical procedures, blood transfusions are imperative for the restoration of lost blood volume. Therefore, it is imperative to ascertain the patient's blood type prior to any blood transfusion. In contemporary applications of blood group determination, the utilization of test serums (Anti-A, Anti-B, Anti-D) that facilitates the precipitation of antigens within the bloodstream has become a prevailing practice. Blood groups are determined by detecting the presence of antigens according to the precipitation of antigens. The observation of precipitation on slides is conducted by the laboratory specialist. However, this observation process can be arduous and time-consuming, requiring sustained attention for extended periods. Consequently, machine vision has emerged as a prevalent tool in contemporary automated approaches, mitigating the expert load required for these processes. Machine vision is a field that is constantly evolving, and one of the most current methods employed is the use of vision transformers. The objective of the present study was to achieve high-accuracy classification of blood groups by employing vision transformers for the purpose of discrimination. The findings of experimental studies have indicated the notable efficacy of vision transformers in the classification of blood groups. Furthermore, experimental findings have indicated that the Adaptive Moment Estimation optimizer when employed in conjunction with vision transformers attains better classification performance.