Department of Computer Science, YSR Engineering College Yogi Vemana University,
Proddatur 516360 Andhra Pradesh, India.

Corresponding author email: rkirans125@gmail.com

Article Publishing History

Received: 30/03/2026

Accepted After Revision: 25/06/2026

ABSTRACT:

The acquisition of clinically reliable blood group data without taking blood samples continues to be a major challenge because traditional approaches are time-consuming, invasive, and rely on laboratory equipment and antigen-based testing. To address these shortcomings, several previous studies have proposed machine learning (ML) and deep learning (DL) methods for determining blood groups from fingerprint images rather than blood samples. Despite the encouraging outcomes of these studies, they have the disadvantages of the low availability of publicly accessible fingerprint datasets and the problem of dataset imbalance. Moreover, numerous earlier studies did not disclose enough transparency on the quantity of images they utilized to educate and appraise, which casts doubts on the dependability and generalizability of the reported accuracies. In order to fill these research gaps, the current study used a publicly available Kaggle standard dataset comprising about 8,000 fingerprint images in eight class labels. Unlike using only the traditional models of ML, including Support Vector Machine (SVM), Random Forest, Logistic Regression, and standard CNN models, the study proposed the DenseNet-201 model that is fine-tuned with a customised fully connected layer with batch normalisation, ReLU activation, and dropout to enhance generalisation. In order to obtain superior regularisation and enhanced generalisation performance, the AdamW optimiser was used in the course of training the model. The proposed DenseNet-201-based model had an accuracy of 94.31. The model allows the prediction of blood groups in real time through fingerprints. Future work will involve incorporating blockchain to store data and using advanced deep learning models to enhance accuracy and robustness.

KEYWORDS:

AdamW Optimiser, Blood Group Prediction, DenseNet-201, Fingerprint Biometrics, Transfer Learning.

Download this article as: Copy the following to cite this article:

Rani K. R, Kiran S, Lakshmi G. S, Vignesh K, Reddaiah S, Prakash K. B. Advanced Fingerprint-Based Blood Group Prediction using Transfer Learning with DenseNet-201. Biosc.Biotech.Res.Comm. 2026;19(2).

Copy the following to cite this URL:

Rani K. R, Kiran S, Lakshmi G. S, Vignesh K, Reddaiah S, Prakash K. B. Advanced Fingerprint-Based Blood Group Prediction using
Transfer Learning with DenseNet-201. Biosc.Biotech.Res.Comm. 2026;19(2). Available from <ahref=”https://shorturl.at/OPOCH“>https://shorturl.at/OPOCH</a>

INTRODUCTION

Blood group detection without blood samples is not easily done because the traditional clinical procedures are time-consuming, invasive, and need laboratory facilities. Traditional methods are antigen-based tests conducted by laboratory technicians to determine the blood group, which is both time-consuming and painful to the patients (Dhande, Bhoir and Gade, 2018). To overcome these weaknesses, several studies have examined ML and DL models to predict blood groups based on fingerprint images, (Pragma et al., 2026).

Deep learning methods are useful in detecting intricate patterns and features like ridges and edge curves existing in the fingerprint images. Image-based classification tasks associated with blood group prediction based on fingerprints using different deep learning and transfer learning models have been implemented, (Prasad et al., 2024).

The use of conventional image processing and ML algorithms like KNN to detect the blood group of a fingerprint was also used earlier (Mounika et al  2024), but later, the use of CNN-based architecture was also implemented to achieve better results, (Vasudev, et al 2025). DenseNet-201 was chosen because it can work with finer-grain fingerprints, and it can reuse features to tackle challenges such as vanishing gradients to enhance accuracy. Transparency and fair comparison were achieved by using a publicly available Kaggle dataset and consistent evaluation, and the model demonstrated a higher accuracy than others and enhanced reliability compared to the past studies. Some of the authors suggested a non-invasive blood group predictor on the basis of fingerprint pictures and various CNN designs. They applied AlexNet, VGG16, ResNet and Inception architectures (Kukadia et al 2020).  The data set was cleaned up and separated into training and test sets. The model obtained an approximation of 92 percent accuracy. They also developed a web-based interface for real-time prediction. However, the study does not distinctly narrate the size of the dataset, the method of validation or cross-validation. Such advanced methods as transfer learning were not elaborated. The convolutional neural network was employed ( Nihar et al  2024).

Prior to the recent popularity of DL, classical machine learning algorithms were utilised. Moslemi et al. (2024) worked on fingerprint images to get ridge features, and the K-Nearest Neighbour (KNN) algorithm was applied to classify them. The technique primarily relies on the type of patterns (loops, whorls and arches). This method is very easy, but it relies on manually extracted features, which does not work well when fingerprint patterns are chaotic or turbulent. An analogous DL method was presented, which used CNN models on blood groups. The analysis points out the advantage of automatic feature learning.

Nonetheless, it fails to compare the outcomes with more powerful pre-trained models and investigate transfer learning policies. It was proposed that it was possible to detect blood group through fingerprint processing based on image processing with CNN classification. The research demonstrates that this is possible because the systems can be non-invasive. Nevertheless, the assessment is insufficient, and such issues as dataset imbalance and cross-population validation are not addressed in an understandable way, (Pallappu et al 2025).

Aamir et al., (2022),  provided statistical relationships between fingerprint patterns and blood groups were analysed, showing meaningful correlations between dermatoglyphic features and blood groups. Further analysis on fingerprint patterns in relation to blood groups and gender was conducted, strengthening the biological association. Ali et al. (2016) discussed foundational fingerprint recognition techniques based on minutiae matching, contributing to feature extraction methodologies used in biometric systems. In (Dong et al., 2017), image-processing-based blood group detection methods were explored, validating the feasibility of automated classification systems.In (Patil and Ingle, 2021), introduced fingerprint map-based approaches for predicting blood groups, focusing on pattern analysis. Workers like  Shrein, (2017), Dong et al (2017), demonstrated  CNN-based fingerprint classification, using ridge orientation images, improving feature learning techniques.

Recently, Agastin et al., (2026) suggested a non-invasive approach to define blood groups based on fingerprint images and a CNN, which learns ridge features. The objective is to substitute the conventional invasive tests with a speedy and cost-effective biometric solution. As the relationship between fingerprints and blood groups is studied, an innovative non-invasive system based on fingerprint images was introduced, emphasising real-world applicability.

Swathi et al  (2024) improved DL based fingerprint classification methods to enhance prediction accuracy. Chelladurai et al.(2024). Enhanced DL architectures for fingerprint-based blood group prediction were proposed, demonstrating better performance and reliability. Genomic-based DL models have demonstrated great accuracy in blood group antigen prediction, but they demand intricate laboratory systems and are not as applicable in biometric systems. The viability of automated blood group prediction, based on simple CNNs, is confirmed by other studies; however, they used simple features. Also, statistical studies show that the pattern of the fingerprint has a biological correlation with the blood group; the loop patterns are more frequent with blood group O.

Current literature indicates that ML and DL are equally capable of matching blood groups based on fingerprints, although much of the literature uses simple CNNs or manual features with little validation. This provides the space to enhance precision and generalisation through state-of-the-art pre-trained models such as DenseNet in an end-to-end method.

MATERIAL AND METHODS

Existing Method: The current approach is the use of a convolutional neural network (CNN)-based approach to detect blood groups on fingerprints (Prasad et al., 2024). To enhance the effectiveness and accuracy of the classification, various CNNs like AlexNet, VGG16, ResNet, and Inception are considered. These models are tested by conventional measures such as accuracy, precision, recall and F1-score. The steps in the methodology are image preprocessing, data augmentation, and model optimisation to improve classification (Sharma et al., 2025). Model reliability is also evaluated by cross-validation. The system architecture is introduced as a conceptual model of how the system looks and how it operates. Moreover, there are new research studies in the field that have used more sophisticated methods of feature extraction, including SIFT, ORB, and Gabor filters as well as CNN-based models, to improve the performance of blood group detection using fingerprints (Sai et al  2025).

But there are some inconsistencies that are noticed in the current approach. Though the dataset is indicated as a processed fingerprint dataset in the past, there are no explicit indications of the size of the dataset and the distribution of classes. Also, there are differences in the values of reported accuracy, where the reported accuracy is about 92-93% in various parts of the study. Although the discussion mentions DL models like VGG16, ResNet, CNN, AlexNet and Inception, in the results section, there are common machine learning classifiers like Random Forest and Decision Tree, which confuses the discussion about the experimental set up and evaluation, (Kumar et al., 2026).

Proposed Method: The primary reason to conduct this approach is that some of the past studies did not include much information regarding their dataset and presented the accuracy findings without a clear description of dataset size, distribution of classes, or validation steps. In order to close these gaps, we proposed the new DenseNet-201 model that is more sophisticated than traditional CNN designs employed in previous blood group prediction systems based on fingerprints,(Tabassum and Mahabubullah, 2025).

In the proposed approach, the study presented a detailed description of the dataset such as the total number of images, the number of images in a specific class and the train-test data split. Unlike using the direct application of DenseNet-201, the study initially trained several deep learning networks like EfficientNet-B1, ResNet-RS-50, and CNN on the same data with the same 80:20 train-test split to provide a fair comparison. Previous studies mostly concentrated on standard CNN architectures without much comparison in terms of architecture. DenseNet-201 was more accurate than the rest of the trained models and better than the current base study, reaching about 94.31% accuracy. In contrast with some earlier literature where the traditional CNN models were primarily used or where optimisation methods were restricted, (Anuradha  et al., 2025).

The proposed DenseNet-201 model was trained using a customized and optimized fashion to avoid underfitting and overfitting. Downsampling, appropriate optimisers (Adam), dropout layers, and hyperparameter learning rate, number of epochs, and weight adjustments were all a part of the optimization process. High-level training methods was also used to enhance training effectiveness and model stability. During training, overfitting was constantly checked, and the parameters of the batch size and learning rate were adjusted. The trained models were tested on standard performance measures which comprise accuracy, precision, recall and F1-score which are commonly used in blood group prediction using fingerprints studies, (Adgaonkar et al., 2025).

The steps involved in the preprocessing of proposed method are discussed below and also represented in Figure 1.

Data Collection:The current study has  suggested a publicly available dataset of fingerprints images that has about 8,000 images of various classes. There are different categories of blood groups associated with each label of the class. The data was well structured, such that the images were well marked and can be utilized to train and evaluate. The general data was split into training and testing in the ratio of 80:20.

Data Preprocessing: Given that the data has certain noisy and blurred images of the fingerprints, preprocessing was conducted to enhance the quality of the images and the model performance. The preprocessing methods comprised of conversion to grayscale and normalization. Moreover, random horizontal flip, random rotation, and random adjustment (sharpness/brightness variations) data augmentation methods were used. These measures will assist to enhance generalization of the model and minimize overfitting.

Feature Extraction:The process of feature extraction was conducted automatically with the help of the convolutional neural network (CNN) implemented as part of the DenseNet-201 architecture. The network automatically identifies significant fingerprint ridges patterns and structural features when the input image is sent through the convolutional layers. This does not require manual feature extraction and the model is able to learn meaningful representations directly with the data.

Training Strategy (DenseNet-201 Model): The DenseNet-201 model was trained on the dataset, separated into the training and testing subsets. The model was adapted and optimized with Adam optimizer and a loss function (Cross-Entropy Loss). In order to minimize overfitting, dropout layers were added. The use of advanced training methods like Automatic Mixed Precision (AMP) was also employed to enhance the computational efficiency and minimize the use of memory leading to training that was faster than that of the conventional methods. Training was performed with hyper parameters like learning rate, number of epochs and batch size being carefully tuned. Continuous monitoring of overfitting was performed and required changes were done to provide consistent and sound performance.

Figure.1 Proposed Processing Method.

Model Evaluation: Several deep learning models were trained and tested using the same dataset in comparison. They were EfficientNet-B1, ResNet-RS-50, a basic CNN model, and DenseNet-201. The models were trained on the same splits of the dataset so that they are fair. The standard performance measures (such as accuracy, precision, recall, and F1-score) were used to evaluate the trained models.

Proposed DenseNet-201 Model Architecture: At this point, the proposed architecture model in more detail and clarify why it assists in attaining a greater accuracy than the existing approaches to predicting blood groups based on fingerprints which is presented in Figure 2. In particular, the proposed model examine the contribution made by the DenseNet-201 model to the increased detection performance. DenseNet-201 architecture boosts feature reuse and gradient flow, and that is why this model can adequately represent a complex ridge pattern that appears in fingerprint images. Therefore, the suggested method enhances the accuracy and reliability of blood grouping by fingerprints.

Figure 2: Denset Net-201 Architecture.

In this DenseNet-201, fingerprint images are first preprocessed before being given as input to the DenseNet-201 model. All the images are resized to 224 x 224 and changed to three channel formatto fit the demand of the pre-trained network. The pixel values are brought to a normalized form to have a stable and quicker convergence in training. Following the preprocessing, the image is subjected to the first convolutional and pooling layers of the DenseNet-201 where simple fingerprint characteristics like the ridge edges, curves, and texture patterns are obtained. As the image goes deeper into the network, dense blocks have a significant role as they are the ones that join each layer to all the earlier layers. This connectivity aids in maintaining the fine ridge details and enhances the feature propagation within the network.Between dense blocks, transition layers are applied to downsample the spatial dimensions of feature maps with important fingerprint characteristics being preserved. The last step is the global average pooling used to map the feature maps extracted to a small feature vector. The feature vector is then submitted to a fully connected layer in which a soft-max transfer is performed to classify the fingerprint to one of the blood group categories.

RESULTS AND DISCUSSION

As it was mentioned above, once the dataset was prepared and the model training process was done, the results of classification in various blood groups were obtained. The standard performance measurements such as accuracy, precision, recall and F1-score were used to evaluate the trained model. The findings show that the proposed model could successfully identify various blood groups with a steady degree of accuracy among the classes.

The results which are shown  in Table.1 indicate that the suggested method works well in practice and can address some of the limitations that are found in the traditional methods applied to detect blood groups. The performance of the proposed model was observed to be better and more reliable as compared to conventional models. The general findings show encouraging performances especially because of the presence of a well-organized data set and precise experimental design, which improves the reproducibility and openness of the study.

Table 1. Results of Proposed Method (DenseNet-201).

Blood

Group

Accuracy Precision Recall F1-Score

score

Support A+ 0.95 0.96 0.95 0.96 199 A- 0.91 0.93 0.90 0.92 217 AB+ 0.96 0.96 0.96 0.96 187 AB- 0.95 0.94 0.94 0.94 206 B+ 0.93 0.97 0.92 0.94 202 B- 0.99 0.90 0.98 0.94 193 O+ 0.92 0.93 0.92 0.92 195 O- 0.94 0.91 0.94 0.92 201 Accuracy – 0.9431 0.9431 0.94 0.94 Macro Avg – 0.9437 0.9438 0.94 1600 Weighted Avg – 0.9437 0.9431 0.94 1600

Evaluations of the model were based on common classification measures, namely, accuracy, precision, recall, and F1-score. The proposed DenseNet-201 model obtained the total accuracy of 94.31, the macro average and weighted average F1-scores of about 94.34 and 94.31, respectively. The similar values of the macro and weighted averages show the balanced performance of the various blood group classes. Analysis by classes indicates that there is a regular detection performance among all categories. As an illustration, the B-class had a high recall, whereas other classes, including AB+, A+, and O- had large values of precision and F1-score.

Even though some differences were evident in some classes like O+ and A-, the general performance was constant and dependable. Although, slight variations were observed among certain classes such as O+ and A−, the overall performance remained stable and reliable. The similarity of the values of precision and recall of the different classes implies that the model is not biased towards any specific blood group. The achieved results reveal that the suggested method is efficient in predicting the blood group based on fingerprints.The results of the performance have shown that DenseNet-201 can extract meaningful ridge-based features of fingerprint images and attains an average classification accuracy with multiple types of blood groups.

Confusion Matrix Analysis: The confusion matrix gives a more detailed analysis of the accuracy of the proposed DenseNet-201 model to each blood group. The correct classification of samples is indicated by the diagonal values of the matrix and the misclassifications are indicated by the off-diagonal values.

Figure 3: Confusion Matrix

Based on the confusion matrix presented in Figure.3, one can see that the greatest number of predictions falls on the diagonal and this proves that the model was able to classify a lot of samples in each category of blood group correctly. As an illustration 190 samples of A+ were correctly identified, 197 samples of A- were correctly classified and 195 samples of AB- were accurately predicted. Likewise, B−(191 accurate predictions) and O−(189 accurate predictions) are also good performers in classification.

There are some small misclassifications that are observed among the closely related blood groups. An example is that some few B+ samples were wrongly labeled as B-, and some O+ samples were wrongly labeled as O-. Equally, small confusion is also witnessed in few cases between A – and B -. These errors are however very minor in comparison to the total sample count in each of the classes.

On the whole, the confusion matrix demonstrates that the proposed model has the balanced performance in all eight blood group classes. The large proportion of correct predictions along the diagonal and the proportionately low number of off-diagonal errors show that the model is valid and there is no strong bias in it to any particular category of blood group.

Comparative Analysis of Different Models: In order to determine the efficiency of the suggested strategy, several deep learning models were trained and compared on the identical dataset which are tabulated in Table 2. The models were a basic CNN model, ResNet-50, EfficientNet-B1 and DenseNet-201. Each of the models was trained in the same experimental conditions to give a fair and unbiased comparison.

Table 2. Comparative analysis of Different Models.

Classifiers Accuracy Precision Recall F1-Score CNN 0.87 0.87 0.86 0.87 ResNet-50 0.92 0.92 0.92 0.93 EfficientNet- B1 0.93 0.93 0.93 0.92 DenseNet- 201 0.94 0.94 0.93 0.94

The standard CNN model, as presented in the Table 2, had an accuracy of 87, a precision of 0.87 and F1-score of 0.87 and recall of 0.86. The CNN model demonstrated rather lower performance in all evaluation metrics as compared to the other models.ResNet-50 was followed by an accuracy of 92 with precision and recall of 0.92 and an F1-score of 0.93. It means that this is much better than the basic CNN model, as it has a higher ability to extract features. The further results were achieved with the help of EfficientNet-B1, which obtained an accuracy of 93, precision, and recall values of 0.93 and 0.92, respectively. The overall accuracy of EfficientNet-B1 was a bit higher than that of ResNet-50. Lastly, DenseNet-201 was the most successful of all the models. It achieved an accuracy of 94, precision, recall and F1-score of around 0.94, 0.93 and 0.94, respectively. The findings show that DenseNet-201 is more balanced and gives more results.

Stable performance in terms of classification as compared to the other models has been shown by Bhadre and   Ghongade (2024). The comparison is valid because all the models were trained and tested on the same dataset and under the same experimental condition. According to the results of observation, in the study, DenseNet-201 seems to be the most appropriate architecture in terms of predicting blood groups based on fingerprints, and the comparison can be trusted and the results can be reproduced.

Figure 4: Assessment of performance metrics across the proposed and state-of-the-art models.

The graphical analysis shown in Figure 4 compares the performance of four classifiers: CNN, ResNet-50, EfficientNet-B1, and DenseNet-201 based on performance indicators like accuracy, precision, recall and F1-score. based on the graph, it can be seen that CNN performs the worst out of the four models with the values of most of the metrics ranging from 0.86 to 0.87. This demonstrates that the model works reasonably and not as good as the other classifiers. The ResNet-50 model is more effective, as the scores of all the metrics are approximately 0.92-0.93.  The ResNet-50 model shows better performance, with scores around 0.92–0.93 for all the metrics.

This implies that it can categorize the data more precisely than CNN. Equally, EfficientNet-B1 has good and balanced scores, and most evaluation metrics are near 1.93.  This implies that the model is consistent in various measures. DenseNet-201 is the most successful overall model with the highest results of about 0.94 in accuracy and precision and slightly higher in recall and F1-score. This indicates that DenseNet-201 is better than the other classifiers. On the whole, the graph indicates that the performance becomes better with CNN to DenseNet-201, and DenseNet-201 provides the best results with the provided dataset.

CONCLUSION

The paper suggests a non-invasive blood group prediction system based on fingerprint images using DenseNet-201 transfer learning. It has high accuracy and reliability and is better than the methods used before that are not transparent. Various models were compared to be validated. The process is reproducible, needs less training effort and energy, and is a time-saving, stable and reliable substitute to conventional methods. The model allows the prediction of blood groups in real time through fingerprints. The future work will involve incorporating blockchain to store data and using advanced deep learning models to enhance accuracy and robustness.

Conflict of interest: Authors do not have any conflict of interest.:

Data Availability: All datasets are available with the corresponding author on reasonable request  

Funding: Nil

REFERENCES

 Aamir, Yasmin & Aamir, Yasmin & Masood, Riffat & Irshad, Nasim & Malik, Dr & Farid, Noreen & Shahab, Muhammad.  Masood, Riffat & Irshad, Nasim & Malik, Dr & Farid, Noreen & Shahab, Muhammad.  (2022) Relationship between Pattern of Fingerprints and Blood Groups,” Pakistan Journal of Medical and Health Sciences, Vol:16 Page:698–700. Available at: https://doi.org/10.53350/pjmhs22169698.

Adgaonkar, Amarja and Vinoth, R. and Raghuveer, K and Misra, Siddharth and Prasad, P.M.K. and Ramamurthy, Madhumitha, AI-Driven Classification and Prediction of Blood Groups through Image Processing (November 15, 2025). Proceedings of the 3rd International Conference on Optimization Techniques in the Field of Engineering (ICOFE-2024), Available at SSRN: https://ssrn.com/abstract=5085490 or http://dx.doi.org/10.2139/ssrn.5085490

Agastin S  , Charles Y  , Mr.S.Kandeeban, Dr.S.Akila. (2026) Deep Learning Based Blood Group Prediction, International Journal for Research in Applied Science & Engineering Technology (IJRASET),  ISSN: 2321-9653, Vol:14 Issue III, Page:1599-1605, Available at:  https://doi.org/10.22214/ijraset.2026.78231.

Ali M M H  Vivek Mahale, Pravin L. Yannawar and Ashok Gaikwad. (2016) Fingerprint Recognition for Person Identification and Verification Based on Minutiae Matching, 2016 IEEE 6th International Conference on Advanced Computing (IACC) in, Page: 332–339. Available at: https://doi.org/10.1109/IACC.2016.69.

Anuradha T.B, Supreet M, Vishwa  and K, Sunil.  (2025) A Non-Invasive Approach for Detection of Blood Group Using Fingerprint Analysis Based on Deep Learning,International Research Journal on Advanced Engineering Hub (IRJAEH), Vol:3, Page:3056–3063. Available at: https://doi.org/10.47392/IRJAEH.2025.0450.

Bhadre  AA and   H P. Ghongade  (2024) Detection Of Blood Groups Through Deep Learning And Image Processing Spvryan’s International Journal of Engineering Sciences & Technology (SEST) ISSN : 2394-0905 Vol:10(3), Page:1–11 Available at: https://spvryan.org/archive/Issue3Volume10/01.pdf

Chelladurai, Sivamurugan & B, Perumal & Siddarthha, Yelchuri & Murthi, Vavilala & Sankar, Yaramati & Varun, Yerra.  (2024) Enhanced Blood Group Prediction with Fingerprint Images using Deep Learning, Conference: 2024 3rd International Conference on Automation, Computing and Renewable Systems (ICACRS) in, Page:1091–1097. Available at: https://doi.org/10.1109/ICACRS62842.2024.10841725.

Dong YF and Wei-wei Fu and Zhe Zhou and Nian Chen and Min Liu and Shi Chen (2017) “ABO blood group detection based on image processing technology,” in 2017 2nd International Conference on Image, Vision and Computing (ICIVC), Page:655–659. Available at: https://doi.org/10.1109/ICIVC.2017.7984637.

Dhande, A., Bhoir, P. and Gade, V. (2018) Identifying the blood group using Image Processing,” International Research Journal of Engineering and Technology, Vol:5(3), Page: 2210-2214. Available at: https://www.ijrte.org/wp-content/uploads/papers/v8i1/A2154058119.pdf

Mounika,G  M. Anusha, D. Gopika, B.S. kumari (2024) Blood group detection through finger print images using image processing (KNN),” International Research Journal of Engineering and Technology (IRJET), Vol:11(03), Page:1225–1228. Available at: https://issuu.com/irjet/docs/irjet-v11i3169.

Kukadiya U, Pratik Trivedi Ashish Rathva, Chintan Lakhani . (2020) Study Of Fingerprint Patterns In Relationship With Blood Group And Gender In Saurashtra Region,” International Journal of Anatomy and Research, Vol:8(2.3), Page:7564–7567. Available at: https://doi.org/10.16965/ijar.2020.159.

Kumar, Kishanjee & Prasad, Rohit & Ghosh, Suman & Mahanta, Sourav.  (2026) Blood group prediction using fingerprint. Research Square Vol:1 Page:1-26,Available at: https://doi.org/10.21203/rs.3.rs-9250543/v1.

Moslemi C, Sækmose S, Larsen R, Brodersen T, Bay JT, Didriksen M, Nielsen KR, Bruun MT, Dowsett J, Dinh KM, Mikkelsen C, Hyvärinen K, Ritari J, Partanen J, Ullum H, Erikstrup C, Ostrowski SR, Olsson ML, Pedersen OB. (2024) A deep learning approach to prediction of blood group antigens from genomic data. Transfusion, Vol:64(11), Page: 2179–2195. Available at: https://doi.org/10.1111/trf.18013.

Mygapula,  SN Bairysetti Prasad Babu, Venkatesh Kondaveti (2025) Deep Learning Based Blood Group Prediction From Fingerprint Analysis, International Journal Of Creative Research Thoughts – IJCRT ISSN 2320–2882. Vol:13(4), Page:j465-j473. Available at: http://www.ijcrt.org/papers/IJCRT25A4119

Pragna R, Hitasmi & N, Jeevan & Khan, Ayan.  (2026) Blood Group Prediction Using Fingerprint Through Deep Learning International Research Journal on Advanced Engineering Hub (IRJAEH), Vol:4, Page:18–26. Available at: https://doi.org/10.47392/IRJAEH.2026.0003.

Nihar, T., Yeswanth, K. and Prabhakar, K. (2024) Blood group determination using fingerprint, MATEC Web of Conferences, Vol:392, Available at: https://doi.org/10.1051/matecconf/202439201069.

Pallapu, S G.Meghana, CH.Madhuri, & N.Sathwika (2025) Crimsonprint: A Deep Learning Framework For Noninvasive Blood Group Prediction From Fingerprint Images, American Journal of AI Cyber Computing Management, Vol:5, Page:382–388. Available at: https://doi.org/10.64751/ajaccm.2025.v5.n4.pp382-388.

Patil N. Vijaykumar and D. R. Ingle (2021) A Novel Approach to Predict Blood Group using Fingerprint Map Reading in, 2021 6th International Conference for Convergence in Technology (I2CT) Page:1–7. Available at: https://doi.org/10.1109/I2CT51068.2021.9418114.

Prasad, Dr & Niharika, K. & Harshitha, D. & Saida, SK. (2024) Blood Group Prediction Using Fingerprint International Research Journal on Advanced Engineering Hub (IRJAEH), Vol:2, Page:2835–2839. Available at: https://doi.org/10.47392/IRJAEH.2024.0392.

Sai, K., Raju, D.N.T. and Vasarao, S. (2025) Fingerprint-Based Pattern Recognition for Accurate ABO Blood Group Prediction International Research Journal on Advanced Engineering Hub (IRJAEH), Vol:3, Page:4001–4005. Available at: https://doi.org/10.47392/IRJAEH.2025.0585.

Sharma, Trapti & Deka, Daisy & Parida, Ayusi & Verma, Shweta & Shukla, Anushree & Islam, Jaabir.  (2025) Fingerprint based Blood Group Detection Using Machine Learning, in 2025 International Conference on Electronics, AI and Computing (EAIC), Page:1–6. Available at: https://doi.org/10.1109/EAIC66483.2025.11101495.

Shrein, J.M. (2017) Fingerprint classification using convolutional neural networks and ridge orientation images in 2017 IEEE Symposium Series on Computational Intelligence (SSCI), Page:1–8. Available at: https://doi.org/10.1109/SSCI.2017.8285375.

Swathi P, Sushmita K and Prof. Kavita V Horadi (2024) Fingerprint Based Blood Group using Deep Learning, International Journal of Advanced Research in Science, Communication and Technology, 4(1) pp. 699–708. Available at: https://doi.org/10.48175/ijarsct-15393.

Sasidhar B, Ashok K, Kishorbabu M, Sameer Sk, Dhavez Parveen Sk, Sai P and Sai Chandu R (2025) “An Innovative Non-Invasive Blood Group Detection Using Fingerprint Images,” Journal of Criminology and Forensic Studies, Vol:7(1), Page: 1-6 Available at: https://doi.org/https://academicstrive.com/JOCFS/articles/view/2006.

Tabassum, M. and Mahabubullah, D.K. (2025) Fingerprint-Based Blood Group Prediction Using Deep Learning. International Journal of Innovative Research in Engineering, Vol:6(4), Page:53–57. Available at: https://doi.org/10.59256/ijire.20250604009.

Vasudev  A. B Biradar D Vaishnavi Janani AM (2025) Fingerprint-based blood group detection using image processing. Intl J of Res. Publication and Reviews  Vol 6 No 1 Pp 5324-5327 Available at: www.ijrpr.com.