International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

Blood Group Detection Using Image Processing and Deep Learning

Jashwanth Sai Ganta Dr. Mohana Roopa Y Mary Rishitha Jaya Surya Pulivarthi
Department Of Computer Department of Computer Department of Computer Department of Computer
science and Engineering science and Engineering science and Engineering science and Engineering
Institute of Aeronautical Institute of Aeronautical Institute of Aeronautical Institute of Aeronautical
Engineering Hyderabad, Engineering Hyderabad, Engineering Hyderabad, Engineering Hyderabad,
Telangana Telangana Telangana Telangana
----------------------------------------------------------------------***-----------------------------------------------------------------------
ABSTRACT— Blood group identification is an procedures, manually evaluating and interpreting blood
essential part of medical diagnostics and transfusion samples is a time-consuming and prone to human error
treatment procedures. Our novel approach to blood type procedure [1]. By combining deep learning and image
detection is presented in this work using deep learning and processing techniques, a novel approach has been
image processing techniques. Orientated FAST, rotated implemented to address these challenges, improving
BRIEF (ORB), and scale-invariant feature transform (SIFT) transfusion management and patient care while
algorithms are used in our method for feature extraction, automating blood type diagnosis [6].
while convolutional neural networks (CNNs) are used for The recommended method uses convolutional
blood group photo classification. The image processing neural networks (CNNs) for blood group photo
pipeline optimizes contrast and minimizes noise in blood categorization in conjunction with feature extraction
group photos using SIFT and ORB before extracting utilizing orientated FAST, rotated BRIEF (ORB), and scale-
discriminative features. Then, utilizing these traits, a CNN invariant feature transform (SIFT) methods [7]. This
model is trained to precisely categorize blood types. As the approach uses an image processing pipeline to preprocess
CNN model is designed to recognize and comprehend the blood group images to improve contrast and minimize
unique patterns associated with different blood types, it noise before using the SIFT and ORB algorithms to extract
might offer dependable and strong prediction. Using discriminative features. Blood type classification that is
extensive trials on a variety of blood group imaging accurate and reliable is made possible by the CNN model
datasets, we evaluate our method. The results demonstrate that is trained using these variables [11].
the effectiveness of our method for identifying blood types, The goal of this research is to enhance
with outstanding classification accuracy and resistance to transfusion management and patient care by speeding up
variations in image quality. Our recommended approach and accurately analyzing blood sample analysis and
could make blood group identification in medical settings improving automated blood type identification [11]. The
easier and allow for rapid, automated blood sample proposed method combines deep learning with feature
analysis. Additionally, merging deep learning with feature extraction techniques in an effort to increase the precision
extraction techniques enhances the effectiveness and and efficacy of blood type prediction in medical settings.
precision of blood type prediction, hence improving The investigations, which employ a wide range of blood
transfusion management and patient care. group imaging datasets, show how adaptable and accurate
the system is at classifying images, even when picture
Keywords—Transfusion therapy, medical quality varies [13].
diagnostics, Scale-invariant feature transform (SIFT), Deep learning methods combined with image
Orientated FAST and Rotated BRIEF (ORB) algorithms, processing technology for blood type determination have
Feature extraction, Convolutional neural networks many potential applications in the medical industry.
(CNN), Preprocessing, Contrast, Discriminative Through automation and increased accuracy in blood type
features, Resilience to changes, Automated blood detection, this technology can accelerate blood transfusion
sample analysis, Patient care, Transfusion processes and perhaps improve patient outcomes [6]. In
management. order to overcome the shortcomings of manual blood
group identification processes, CNNs are coupled with
feature extraction algorithms to provide a more reliable
I. INTRODUCTION and effective solution.[9]
The proposed technique may preprocess blood
Deep learning and image processing are two
group images, extract discriminative features, and train a
examples of cutting-edge technologies that have CNN model, which makes blood group detection
revolutionized the area of blood group identification and comprehensive and creative. The focus on improving
offer a practical means of enhancing and automating blood type prediction efficiency and accuracy through
blood type detection. In conventional blood group extensive testing on many blood group imaging datasets
identification highlights the method's

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 97
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

adaptability and robustness. The blood group type is to the changing field of blood analysis techniques is
predicted using the blood group detection method in this highlighted in the presentation given at the IEEE
manner [11]. convention [4].
In summary, combining deep learning with image Proposed semiautomated system for “Detecting
processing techniques to determine blood type represents blood group which can be identified by microscopic
a significant advancement in medical diagnostics. The coloring image”. It first preprocesses images by converting
proposed approach offers a workable way to automate the RBC to HIS by color correction, histogram equalization,
and increase blood type identification accuracy, hence and color space conversion. The cumulative histogram will
enhancing patient care and transfusion management. be used to identify the color and texture of the image, and
Because of its innovative technological integration and SVM will be used to assess the matching individual's blood
extensive experimental validation, this technique has the group. It takes a lot of work and requires more
potential to fundamentally alter blood group identification experienced individuals to operate this system [6].
in medical contexts [14]. The Existing work discussed offers a fresh and
creative method for utilizing fingerprint patterns to
predict blood types. In order to make the most of the
II. RELATED WORK
widely used method of finger print identification for blood
The paper titled "A Complete Blood Typing group detection, the study will emphasise the uniqueness
Device for Automatic Agglutination Detection Based on and consistency of fingerprint patterns when combined
Absorption Spectrophotometry," authored by Jose with blood slide analysis [2]. By analysing the little feature
Fernandes, Sara Pimenta, Filomena O. Soares, and Graca patterns that fingerprints provide for unique
Minas, was published in the IEEE Transactions on identification, the study looks into the prospect of using
Instrumentation and Measurement in 2014. This paper fingerprints to identify blood types. This method extracts
describes an automated blood type system that uses geographical data and estimates ridge frequency by using
absorption spectrophotometry to detect agglutination. Gabor filters, which leads to efficient fingerprint matching
Improved blood analysis procedures lead to more accurate [8].
and efficient outcomes thanks to this creative method. The The study also provides a historical overview of
study conducted by the authors for this IEEE paper fingerprint-based identification systems, emphasizing
examines developments in measurement and their significance for a variety of applications. This
instrumentation methods for blood type applications, background information provides valuable insight into the
making a significant addition to the area [1]. evolution of fingerprint-based identification and its
The paper authored by Dr. D. Siva Sundara Raja applicability in the modern era. Blood group
and J. Abinaya, was published in the International Journal determination and fingerprint patterns together represent
of Advance Study and Research Work in March 2019. This a novel intersection of biometric data and medical
study introduces an economical approach for blood group diagnosis, with potential implications for patient
identification by employing fingerprints. The method identification and customized medication [12].
presents a potentially resource-efficient solution for The proposed technology has the potential to
determining blood groups, offering a practical application revolutionise blood group detection by offering a reliable,
in healthcare settings. The research, documented in this non intrusive method that leverages fingerprint patterns
2019 journal publication, signifies a noteworthy
inherent distinctiveness [10]. This paper provides an
contribution to advancing cost-effective techniques in the
field of blood group detection, demonstrating the authors' interdisciplinary approach that enables the coexistence of
commitment to innovative and accessible healthcare biometric technology and healthcare through its
solutions [2]. investigation of fingerprint-based identification systems
The suggested technique for “blood group and their potential in medical diagnostics.
identification using image processing is predicated on a
simple methodology”. The technology uses an image of
blood to calculate the standard deviation. If there is an Blood
agglutination reaction and the standard deviation score is Sample
Slide
more than 20, the test is deemed positive. On the other
hand, if the standard deviation result is 20 or less, the test
is considered negative [3]. The agglutination reaction is
crucial for blood type identification, and this method
appears to be a simple and efficient way to determine if it Fig 1: Data Flow diagram of the existing method
is present. By calculating standard deviation, the method
looks for alterations in the blood sample that might
indicate agglutination [5].
This research, documented in the conference IV. PROBLEM IDENTIFICATION
proceedings spanning pages 133 to 136, focuses on
utilizing spectrophotometry for the analysis of RH The problem description pertains to the
phenotypes in human blood typing. The authors' research requirement for a technique that is accessible, affordable,
shows a development in the use of spectrophotometric and rapid to ascertain a person's blood type. Conventional
methods to improve blood type accuracy, especially in blood typing techniques frequently call for specialized
differentiating between RH genotypes. The importance of labs and drawn-out
their contributions

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 98
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

processes, which can be problematic in emergency

situations or places with limited resources. Additionally,
the current approach, which involves intrusion, is less
effective and necessitates the services of a professional.
The project's goal is to create a non-intrusive technology
that uses spectroscopic pictures to precisely forecast a
person's blood grouping in order to overcome this,
offering a practical and effective substitute for current and
conventional blood typing techniques. The spectrometer
work displayed here is very efficient for image processing
chores, including image to binary conversion and thinning
for spectroscopic image correction and normalization for
the prediction.
Fig 2: System Mechanism for Proposed Method
V. PROPOSED FRAMEWORK
The software we utilized is Visual Studio Code,
The recommended method for blood group identification and the Blood Group Detection System dataset is publicly
is a noteworthy advancement in the field of medical available on Kaggle.com. The simplified code editor Visual
diagnostics. By merging deep learning and image Studio Code supports version control, task execution, and
processing methods, the system leverages computer vision debugging. The goal is to provide developers with the
and machine learning to automate the blood group minimum set of tools necessary for a brief code-
diagnosis process. Complex features are extracted from builddebugging cycle, more complex operations are left to
blood samples using convolutional neural networks IDEs with greater functionality, such as the Visual Studio
(CNNs). Because of this, the system can now analyse and IDE. Python is the main language used for machine
identify minute patterns and characteristics that may be learning and deep learning in this project, and it may be
suggestive of specific blood types. Additionally, the built on Windows or Linux (Ubuntu, CentOS). Data
addition of Scale-Invariant Feature Transform (SIFT), handling and visualization will need the use of Python
Oriented FAST, and Rotated BRIEF (ORB) algorithms modules like NumPy, Pandas, Matplotlib, and Scikit-learn.
enhances the system's ability to recognize and match The front-end web developer will work with HTML and
significant points and descriptors within the images, CSS. Libraries like TensorFlow, Future, Argparse, Hashlib,
contributing to the accuracy and robustness of the blood sys, and tarfile will make deep learning easier. Integrated
group classification process. development environments (IDEs) like PyCharm, VS-Code,
or jupyter Notebook are useful for developing and
Though this multi-phase procedure appears highly experimenting with code.
promising, it is important to consider possible
disadvantages, such as the need for large and diverse
training datasets to ensure the system's accuracy across
variations in blood samples. The computational
VII. EXPERIMENTAL SETUP AND
complexity of deep learning techniques may also need a IMPLEMENTATION
significant amount of processing resources. To ensure the Data Acquisition and Preprocessing, the
system's dependability in real-world applications, it will procedure for predicting a blood sample's blood group
be imperative to evaluate its performance under various using spectroscopic images. Blood types may be identified
conditions, including variations in illumination and picture thanks to these images, which capture important
quality. information including antibodies and antigens. Once high-
resolution images are

Fig 3: Architecture diagram of proposed system

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 99
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

acquired, they undergo preprocessing using methods such which enables data to flow from feature extraction to
as image normalisation and histogram equalisation to blood group prediction. Health care providers may also
standardize image quality, lower noise, and improve enter spectroscopic pictures and receive precise blood
contrast. The accuracy and dependability of the obtained group predictions with an easy-to-use interface. Medical
attributes for blood type prediction depend heavily on this professionals' experience with the system's blood group
preprocessing. Assuming the photos include high-quality, prediction features will be improved by the user-friendly
standardised data, the data collection and preparation interface, which attempts to increase the usage of blood
stages are crucial for getting the images ready for feature group identification technology in healthcare settings and
extraction and blood type prediction using deep learning make it more useful for them to utilize in their job.
algorithms. Testing and Deployment, extensive testing is
Feature Extraction, Utilising the Scale-Invariant conducted on a variety of blood sample types to confirm
Feature Transform (SIFT), Oriented FAST, and Rotated the efficacy of the blood group identification method
BRIEF (ORB) algorithms, features are extracted from utilizing spectroscopic pictures. The system is
preprocessed spectroscopic images in order to identify implemented in healthcare and research environments
blood groups. In order to identify distinctive features that when testing is completed successfully, guaranteeing a
are independent of size and rotation—which are essential smooth interface with the current workflows and
for classifying blood types—these methods are used. The information systems in the laboratory. In order to
ORB highlights the requirement for oriented and establish trust in the system's capacity to provide accurate
differentiating traits, whereas SIFT finds and extracts blood group predictions across a range of datasets,
significant locations and descriptors. The critical extensive testing is conducted to validate the system's
components of the blood samples are intended to be durability and reliability in real environments. Due to its
captured by the extracted features in order to determine interaction with laboratory information systems, the
blood type. These properties are then used as the basis for system becomes a crucial part of laboratory workflows
further analysis and categorization utilising deep learning when it is put into place. It provides useful support for
techniques. blood group analysis and prediction in clinical and
Convolutional Neural Network (CNN) Model research contexts. Through the use of non-intrusive
Training, a Convolutional Neural Network (CNN) model spectroscopic imaging, the general structure is therefore
optimized for blood group prediction in the blood type followed in this order to forecast the blood group.
detection system is developed and trained. In order to
recognize and classify patterns and qualities that belong to
different blood types, the CNN model is engineered to
process and analyze the distinctive properties that are
retrieved from preprocessed spectroscopic photos in an
effective manner. To enhance its ability to accurately
classify blood types, the model is trained using labelled
spectroscopic images that are associated with
corresponding blood group classifications. This helps the
model learn to connect features with blood group
classifications. By utilizing image processing and deep
learning techniques, the trained CNN model seeks to offer
a trustworthy and efficient method of predicting blood
types from spectroscopic images.
Model Evaluation and Validation, the trained CNN
model's accuracy, precision, and generalization ability are Fig 4: Flow chart for Proposed Method
assessed using separate validation datasets during the
blood group detection system's assessment and validation
phase. By comparing the model's performance to
established ground truth labels, measures including
accuracy, recall, and F1 score are used to evaluate how
well the model predicts blood types from spectroscopic
images. In order to maximize prediction accuracy and
generalization to new data, the model's architecture and
parameters are modified based on the evaluation results.
By means of iterative assessment and modification, the
system seeks to guarantee that the CNN model can predict
blood groups with accuracy and reliability. This
methodology provides a stable and efficient blood type
prediction method.
System Integration and User Interface
Development, the blood group identification system's last
phase integrates the trained CNN model with feature
Fig 5: Training and validation data of model
extraction techniques to produce a coherent system. A
fluid and efficient process for spectroscopic image analysis
is provided by this integration,

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 100
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

VIII. RESULTS AND DISCUSSION

The trained convolutional neural network (CNN)
model performed consistently when predicting blood
types from spectroscopic images. The model successfully
classified blood samples into the appropriate categories
using the validation dataset, as evidenced by its overall
accuracy of 95%. The accuracy, recall, and F1-score Fig 7: Training dataset of O +VE
metrics were used to further test the model's effectiveness
in handling different blood sample types and correctly
predicting respective blood groups.
A feature extraction technique using the Scale-
Invariant Feature Transform (SIFT), Oriented FAST, and
Rotated BRIEF (ORB) algorithms was successfully applied
to the spectroscopic images in order to extract distinctive
visual patterns and features. By providing meaningful Fig 8: Training dataset of A +VE
information on the unique qualities of blood samples, the
returned features improved the interpretability of the
predictions made by the CNN model. The interpretability
of the model is crucial in comprehending the decision-
making process and fostering confidence in its predictions
among scholars and healthcare professionals.
Since the blood group detection system provides
a non-intrusive and effective means of predicting blood Fig 9: Training dataset of AB +VE
groups from spectroscopic pictures, its successful
development and validation have substantial therapeutic
importance. Improved patient care and more accurate
diagnoses might result from the system's ability to
transform blood type procedures in medical settings. In
the future, the system might be improved to anticipate
more blood-related factors and integrated with electronic
health record systems for complete patient information
management. Fig 10: Training dataset of B -VE

Live Results of the proposed methodology:

Fig 6: Plotting Testing and Training Accuracy

Fig 11: Live results with dataset of O +VE

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 101
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

IX. CONCLUSION AND FUTURE SCOPE

The development of the blood group
identification system which blends deep learning and
image processing technologies significant advancements
in non-intrusive blood typing methods have been made.
The method has shown to be extremely accurate and
therapeutically useful for determining blood types from
spectroscopic images. Robust feature extraction was
accomplished by using the SIFT (Scale-Invariant Feature
Transform), Oriented FAST, and Rotated BRIEF (ORB)
methods, and convolutional neural networks (CNNs) were
effectively trained. The use of these state-of-the-art
technologies demonstrates the system's effectiveness in
Fig 12: Live results with dataset of A +VE
medical diagnostics by offering a successful and non-
intrusive blood typing approach.
The future scopes are, its intuitive user interface
and seamless integration with existing laboratory
information systems further augment the system's
suitability for medical professionals. The system's
intuitive design, which provides a seamless contact
experience, makes it easy for healthcare professionals to
utilize. Effective data management and workflow
integration are further made possible by the contact with
laboratory information systems, which ensures
compatibility and interoperability with the present
healthcare infrastructure.
Fig 13: Live results with dataset of AB +VE
Potential avenues for future system development
might include the prediction of additional blood-related
traits such as antibody screening and the Rh factor. These
additional components will allow the system to develop
into a more complete blood analysis tool for use in
medical settings. With this development, medical
personnel would be able to understand blood samples on
a deeper and more comprehensive level, improving
clinical decision-making and improving patient outcomes.
The potential of blood group detection
technology is great if it can be simply integrated with
electronic health record (EHR) systems. Through EHR
system integration, the device may automatically add
blood group forecasts to patient records, enhancing
Fig 14: Live results with dataset of B -VE clinical decision-making and patient information
management. Because of this integration, medical staff
members may access and update patient information
more quickly, giving them meaningful knowledge about
patients' blood types in connection to their whole medical
histories.

The predictions of the CNN model require more

research in order to make them more understandable and
interpretable. To increase the confidence that academics
and medical professionals have in the system's
projections, more openness and understanding of the
model's decision-making process should be encouraged.
Improved system-human specialist collaboration might
result in better healthcare decisions and useful research
Table 1: Training Datasets of blood groups discoveries in the long run, thanks to this improved
interpretability.
Future advancements might greatly boost the
blood group identification system's usefulness and
influence in medical research. Researchers and medical
experts may find

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 102
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024 www.irjet.net p-ISSN: 2395-0072

the system to be an even more valuable tool in the future. [10] S. M. Nazia Fathima, “Classification of Blood Types
In addition to enhancing patient treatment, these by Microscopic Color Images”, International Journal of
advancements would increase our understanding of Machine Learning and Computing. Vol. 3, No 4, August
blood- related diseases and traits. The system may 2013.
accomplish these objectives by emphasizing, explainability
[11] Mehedi Talukdar, Md Rabiul Islam, Md. Mahfuz
and interpretability, increasing the range of blood-related
Reza, Mahbuba Begum, Md. Mahmudul Hasan,
factors it can predict, and connecting with EHR systems
“Improvement of Accuracy of Human Blood Groups
with ease.
Determination using Image processing Techniques”,
International Journal of Advanced Research in Computer
REFERENCES and Communication Engineering. Vol. 4 Issue 10, October
[1] Jose Fernandes, Sara Pimenta, Filomena O. Soares 2015.
and Graca Minas, “A Complete Blood Typing Device for
[12] Melur K. Ramasubramanian & Stewart P.
Automatic Agglutination Detection Based on Absorption
Alexander, “An integrated fiberoptic-microfluidic device
Spectrophotometry”, IEEE Transactions on
for agglutination detection and blood typing” Biomed
Instrumentation and Measurement, [2014]
Microdevices, Sept. 2009, pp. 217-229
[2] Dr. D. Siva Sundara Raja and J. Abinaya, “A Cost-
[13] Keerthana. D, Dr. Ranganathan. L “Design and
Effective Method for Blood Group Detection Using
Development of Blood Sample Analyzer Using Intelligent
Fingerprints”, International Journal of Advance Study and
Machine Vision Techniques”, International Journal of
Research Work, Volume 2, [March 2019]
Engineering Science and Computing, Volume 7, pp.
[3] S. Pimenta, G. Minas, and F. O. Soares,
1201712020, May 2017
“Spectrophotometric approach for automatic human
blood typing,” in Proc. IEEE 2nd Portuguese Meeting, [14] Patil, S.M. (2015) “Predictive analytics in higher
[2012], pp. 101-104. education. International Journal of Advanced Research in
Computer and Communication Engineering”, 4(12), 138-
[4] J. M. Fernandes, F. O. Soares, and G. Minas, “RH
141. Doi: 10.17148/IJARCCE.2015.4123.
phenotypes analysis by spectrophotometry in human
blood typing,” in Proc. IEEE 3rd Portuguese Meeting,
[2013], pp. 133-136.
[5] Anurag Sadashiv Phad, Tejas Sanjay Targhale,
Bharat Bhalshankar, Sunita Kulkurni, “Blood Group
Detection by Using Raspberry Pi-3”, International Journal
of Advanced Research in Electronics and Communication
Engineering (IJARECE). Vol 7, Issue 4, [April 2018].
[6] Tejaswini H V, M S Mallikaijuna Swamy
“Determination and Classification of Blood Types using
Image Processing Techniques”, ITSI Transactions on
Electrical and Electronics Engineering (ITSI TEEE),
Volume -2, pp. 2320 - 8945, [2014]
[7] Mansi K, Hitashree M., Chandana Lakshman Hegde
paper, “Blood Group Determination through Medical
Image Processing”, International Journal for Research in
Applied Science and Engineering Technology. Vol 9, Issue
v, [May 2021].
[8] Nuha Odeh, Anas Toma, Falah Mohammed, “An
Efficient system for Automatic Blood Type Determination
Based on image Matching Techniques”, MDPI stays neutral
with regard to jurisdictional claims in published maps and
institutional affiliations. Vol 11, June 2021.
[9] A.Narkis Banu, V.Kalpana, “An Automatic System to
Detect Human Blood Group of Many Individuals in a
Parellel Manner using Image Processing”, International
Journal of Pure and Applied Mathematics, vol-118, pp.
3119-3127, 2018

© 2024, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 103