AI and Simple Cameras Team Up to Automate Human Blood Type Identification

Figure Ilustration AI

FORMOSA NEWS - Sulawesi Barat - Identifying human blood types no longer requires exclusive reliance on manual, error-prone visual analysis by laboratory technicians. A team of computer science researchers from Universitas Sulawesi Barat and Universitas Trunojoyo has developed an automated blood type identification system that utilizes digital image processing and artificial intelligence (AI) to recognize blood types through standard camera images. Published in the Formosa Journal of Computer and Information Science (FJCIS) in 2026, the study introduces a low-cost, software-driven framework designed to accelerate medical diagnoses and minimize human error in clinical settings. The breakthrough system, created by Indonesian researchers Muzaki, Aeri Rachmad, Indra, and Irfan AP, bridges the gap between complex medical procedures and affordable digital technology.

The Challenge of High-Cost Medical Diagnostics
Determining an individual's blood type (A, B, AB, or O) is a critical procedure in emergency medicine, blood transfusions, and prenatal care. While modern automated systems exist, most current innovations rely heavily on high-resolution microscopic imaging and complex computer hardware. These expensive setups create significant barriers for small clinics, rural health centers, and educational laboratories with limited financial resources. To address this technological and economic divide, the research team from Universitas Sulawesi Barat and Universitas Trunojoyo focused on designing an efficient system that works with affordable, everyday equipment. By replacing expensive laboratory hardware with standard digital cameras and lightweight computing algorithms, the authors created a highly adaptive diagnostic model suited for resource-constrained environments.

Methodology: From Blood Clots to Binary Matrices
The automated system mimics the traditional manual testing method, which observes how blood reacts when mixed with specific antigen reagents (anti-A, anti-B, and anti-AB). Biochemically, these reagents cause blood cells to clump together a process known as clotting or agglutination. To automate this observation without human bias, the system developed by Muzaki and his colleagues processes images through a highly structured, multi-stage software pipeline:

• Image Acquisition: A standard handycam camera captures the blood samples on a testing slide. A video blaster card then converts these analog visuals into digital images with a resolution of $512\times256$ pixels in grayscale.
  
• Image Preprocessing (Edge Detection): The software applies a mathematical tool called the Prewitt Operator. This technique identifies significant changes in pixel intensity to map out the exact boundaries and outlines of the blood clots, successfully stripping away unnecessary background visual noise.
  
• Feature Extraction: The system divides the cleaned image into an automated grid, scanning the pixels to generate a simplified $9\times10$ binary matrix. Areas with dense blood clotting are assigned a logic value of 1, while smooth, non-clotted areas receive a 0.
  
• AI Classification: This numerical grid is fed into an Artificial Neural Network (ANN) trained with the Backpropagation learning algorithm. The neural network processes the binary pattern through 90 input neurons, 10 hidden layers, and an output layer to deliver the final blood type verdict.
  

Key Findings and System Accuracy
The empirical tests conducted by the research team demonstrate that the combination of basic image processing and artificial neural networks is highly capable of interpreting medical visual patterns. The primary results from the system evaluation include:

• 80% Total Accuracy: In direct testing trials, the AI software successfully recognized and classified 16 out of 20 blood sample data patterns correctly.
  
• Efficient Mathematical Scaling: Representing blood images as a $9\times10$ binary matrix successfully decreased computational complexity, allowing the software to run rapidly without overloading basic computers.
  
• Effective Error Reduction: The backpropagation training graph proved that as the algorithm went through more iterative learning cycles, the system's margin of error dropped significantly, validating the stability of the neural network.
  

The researchers noted that the four instances of misclassification were primarily caused by external environmental factors rather than internal software flaws. Minor variations in room lighting, the physical angle of the digital camera, and the performance limitations of the video blaster hardware occasionally obscured the delicate edge details of the blood clots.

Real-World Impacts and Future Policies
The successful development of this system by the researchers at Universitas Sulawesi Barat and Universitas Trunojoyo carries substantial practical implications for global healthcare equity. By demonstrating that reliable medical classification does not require elite computing setups, this study opens the door for low-cost diagnostic software installations in community health centers and remote regions. Lead researcher Muzaki, representing Universitas Sulawesi Barat, emphasizes the adaptive nature of the project, noting that the combination of simple preprocessing methods and binary representations proves that diagnostic systems do not necessarily require complex, expensive computations to produce adequate medical classifications. To elevate the system for widespread industrial use, the authors recommend expanding the research by utilizing larger image datasets and transitioning to advanced deep learning frameworks, such as Convolutional Neural Networks (CNN). They also envision integrating the software with mobile devices and Internet of Things (IoT) platforms, which could soon allow field doctors to identify blood types instantly using a smartphone camera.

Author Profiles
Muzaki, S.Kom., M.T. Lecturer and researcher at the Department of Informatics, Universitas Sulawesi Barat. His field of expertise focuses on Artificial Intelligence, Digital Image Processing, and Neural Networks.
Aeri Rachmad, S.T., M.T. Academic researcher affiliated with Universitas Trunojoyo. He specializes in Computer Vision, Pattern Recognition, and Software Engineering.
Indra Computational researcher at Universitas Sulawesi Barat, specializing in digital systems and information technology development.
Irfan AP Technology researcher at Universitas Sulawesi Barat, focused on software implementation and data processing applications.

Source
Muzaki, Aeri Rachmad, Indra, Irfan AP. Blood Type Identification System in Humans Based on Digital Image Processing. Formosa Journal of Computer and Information Science (FJCIS)
DOI: https://doi.org/10.55927/fjcis.v5i1.16641

URL: https://journal.formosapublisher.org/index.php/fjcis