Lifeline : a portable blood group identifier using image processing and machine learning for mobile blood-donations. 6

Undergraduate Thesis: (BS in Computer Engineering) - Pamantasan ng Lungsod ng Maynila, 2024. 56

5

ABSTRACT: STATEMENT OF THE PROBLEM: Blood mismatch in transfussions, often due to clerical errors, can be fatal. Mobile donation drives conduct manual screening of blood groups by medical technologies. This reliance on human judgement increases the risk of errors. The proponents aim to develop a portable device for automated blood group identification which improveds the process of identifying blood groups from samples, optimizes information organization, and ultimately, to eradicate the risk for blood mismatch. Specifically, this study to attain the following objectives: 1. Develop a portable blood group identification device for mobile blood donations. 2. Develop a CNN-based model that can be correctly identify blood groups with 100% match rate. 3. Build a database that will serve as a repository for the donor's personal information, the images captured during the blood type identification process, and the practitioner conducting the blood typing process. RESEARCH METHDOLOGY: The research design of this study centers around a primarily quantitative approach, concentrating on evaluating the accuracy of the CNN-based model by conducting a comparative analysis between the match results between the traditional manual blood grouping and lifeline's automated blood grouping. The study utilizes advanced image analysis techniques and machine learning algorithms for automated blood group identification. Data is obtained through controlled experiments, and a model validation employed to quantify the performance metrics (precision and recall) of the automated system. SUMMARY OF FINDINGS: LIFELINE provided a great assistance during mobile blood donatios with its portable design, enabling on-site utilization for blood group determination. During its model validation, the model showcased an impressive average precision of 97.6% across all blood classes, indicating high accuracy in positive predictions. Similarly, maintaining an average recall of 100% undersores of the model's ability to capture all instances of each blood type. In its model accuracy testing, lifeline was able to achieve a perfect 100% accuracy rate in comparison to the medical practitioners identified blood group. Furthermore, LIFELINE's website effectively stores essential donor information including name, age, sex, identified blood group, acquisition date, acquiring practitioner, and original blood sample images, blood sample image. CONCLUSION: The sytem was successfully developed in identifying blood groups using image processing and machine learning through CNN-based Model with 100% match accuracy while also being able to store significant donor information. This study proved that it is an effective and convenient addition in Mobile Blood Donation Drives. The proposed method of using image processing techniques enables ways to improved the machine learning model, thus making it more accurate as a blood group identifier. The predictions shown as a result of testing were also quick and accurate. The solutions innovated by Lifeline show immense potential in various applications in the medical field through the utilization of machine learning. RECOMMENDATIONS: The researchers suggest several improvements for future studies. First, including the grade of agglutination would boost confidence in identifying blood groups. Second expanding the dataset for the rare AB blood group is essential to enhance Lifeline's model accuracy. Third, increasing the dataset size for dilute mixture samples would refine Lifeline's model to accurately identify blood types, especially in scenarios with limited blood volume. Lastly, optimizing the dimensions of the device by reducing its length and widtch would improve portability, making it more convenient for healthcare professionals during mobile blood donation drives.

5