Deep Learning Approaches for Anemia Diagnosis through

Classification Techniques

A. Deenu Mol

, S. Subashini

, S. Karthikkumar

, P. Hrithikkumar

, K. Mohammed Ashraf

and S. Kavin Prabhu

Department of Information Technology, Kongu Engineering College, Perundurai,Erode 638060, Tamil Nadu, India

Department of Electrical and Electronics Engineering, Jai Shriram Engineering College,

Tirupur – 638660, Tamil Nadu, India

Keywords: Anemia Detection, Fingernail Images, Inception v3, Support Vector Machine (SVM), Random Forest, Deep

Learning, Machine Learning.

Abstract: A non-invasive method for anemia detection via nail images was developed in this study based on deep

learning and machine learning approaches. Data were acquired and preprocessed in this study, including

augmentation and normalization to improve model performance. Feature extraction was conducted with the

Inception v3 model and then classified using the Random Forest and Support Vector Machine algorithms.

This is an effective way to predict anemia with less cost and time than conventional blood tests. The

performance evaluation was done by using accuracy and confusion matrix in which promising results were

achieved in detecting non-invasive anemia. The combination of deep learning with Random Forest and SVM

gives a scalable solution with the most advantages in resource-poor areas.

1 INTRODUCTION

Anemia is a common blood condition that

corresponds to lower amounts of red blood cells or

hemoglobin and consequently less transport of

oxygen to various body parts. However, the

traditional methods of diagnosing anemia rely on

tests, such as complete blood counts and peripheral

blood smears, which need blood samples and may not

be accessible to certain population groups because of

the costs involved or resources available. Therefore,

there is increasing interest in alternative, easier, and

more feasible types of diagnostics.

The method investigated in this research is

fingernail image processing for anemia detection,

which actually exploits all possible advantages

brought by deep learning and machine learning

techniques to facilitate their application. Fingernails

are easily accessed and serve as the body's mirror

reflecting the skin paleness due to anemia, thus

offering a possible opportunity to develop visual

input for anemia detection. In this case, feature

extraction was done using the Deep Learning

framework Inception v3 model, recognized to have a

strong feature extraction capability. These features

were subsequently classified using the Random

Forest and Support Vector Machine algorithms so

that their performance in predicting anemia could be

evaluated and compared. Data preprocessing by

means of data augmentation was applied so as to

enhance the robustness and generalizability of the

models, including such techniques as rotations and

zooming.

This information also helped assess better

performance and reduced overfitting on unseen data.

The combination of deep learning with Random

Forest and SVM proposed in this study represents an

efficient way to identify the visual clues of anemia

and provides a scalable technique for diagnosis that

may be applied to different settings. This research

aims at contributing to machine learning-based

diagnostic tools that may support anemia diagnosis in

a more accessible way across healthcare systems.

826

Mol, A. D., Subashini, S., Karthikkumar, S., Hrithikkumar, P., Ashraf, K. M. and Prabhu, S. K.

Deep Learning Approaches for Anemia Diagnosis through Classiﬁcation Techniques.

DOI: 10.5220/0013921400004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

826-831

ISBN: 978-989-758-777-1

2 RELATED WORKS

Muljono et al. used AI to detect anemia

noninvasively by ways of deep learning, achieving

higher diagnostic accuracy. We extend that further by

taking from InceptionV3 to extract features and apply

machine learning tools for classification. During

2019, Amruthamsh A et al. presented the application

of EfficientNet models for the nonmartial detection of

anemia with image palmar feature extraction, initially

verified in clinical settings .A revolutionary model

combining machine learning models and attention

mechanisms for anemia detection was proposed by

Robert G. Mannino et al. in 2018 .

Using the image processing and the image

thresholding approach, Azwad Tamir and co-workers

(2019) established noninvasive images for observing

conjunctivas, a rapid screening cost-effective method

.Endah Purwanti (2023) trained CNNs on observation

of palpebral conjunctiva images to detect color and

texture patterns for rapid diagnosis .Viveha et al.

developed a point-of-care smartphone application for

the early detection of anemia, thereby enhancing

accessibility to overworked areas and early diagnosis.

A mobile application for anemia screening using

ocular conjunctiva images was developed by Meileth

Rivero-Palacio et al. as a fast and noninvasive tool

offering the means of detection using machine

learning approaches. With applications of

smartphones and image processing, Selim Suner et al.

created the usable tool as a means of hemoglobin

level and anemia risk screening. Justice William

Asare et al. (2023) examined machine learning

algorithms to find which models best help identify

anemia using medical imaging. With palm pallor

assessment as a foundation, Sumana Naskar et al.

(2021) developed a non-invasive test for detecting

anemia, where image processing identifies color

variations .

VR Ravi et al. (2020) evaluated models based on

deep learning in attempts to estimate anemia from

conjunctival images to extract the most potent among

non-invasive screening techniques. Shaun Collings et

al. (2016) assessed multi-class classification

algorithms for anemia diagnosis in clinical settings in

order to augment the precision of diagnosis. Prakriti

Dhakal et al. (2023) explored different machine

learning algorithms to predict anemia, identifying

those which help in the diagnosis as well as the early

assessment and intervention.

Shekhar Mahmud et al. (2023) identified and

implemented non-invasive anemia detection using

CNNs and transfer learning from lip mucosa images,

thereby bolstering non-invasive diagnosis in

resource-poor areas. Rajan Vohra et al. (2022)

proposed multi-class classification techniques aimed

at optimizing anemia detection and bettering

outpatient anemia management. Krithika S et al.

(2023) proposed a multi-input deep neural network

framework to detect anemia non-invasively through

the fingernail images to predict hemoglobin levels,

based on color and texture features. Tariq Ahamed et

al. (2023) created an AI smartphone application

assessing fingernail images through CNNs, aimed at

predicting any possible onset of anemia and thereby

facilitating the diagnosis in areas remote and cursed

with diseases.

Mikhail Ivanov et al. (2020) examined various

preprocessing algorithms on images to aid feature

extraction for noninvasive anemia detection using

deep-learning models. Anika Sharma et al. (2020)

studied how transfer learning with VGG-16 performs

in anemia classification, showing that it can work in

small data sets. Dinesh Reddy et al. (2021) proposed

an automated system for detecting anemia using HSV

color space transformation and k-means clustering on

conjunctival images.

Jose Martinez et al. (2019) examined image-based

hemoglobin estimation methods using deep neural

networks that illustrate how the intensity of pallor

relates to severity of anemia. Felipe Rocha et al.

(2024) proposed a hybrid model that combines CNNs

and transformers to improve anemia prediction

accuracy in low-light smartphone images. Arvind

Kumar et al. (2023) introduced an ensemble deep

learning architecture that incorporated CNN and

LSTM networks, achieving higher accuracy in

anemia detection through conjunctival images.

Sneha Mehta et al. (2021) devised an ensemble

model combining InceptionV3 and XGBoost for

better diagnostics of anemia in clinical scenarios.

Hiroshi Tanaka et al. (2022) adopted an optimized

InceptionV3 model to process lip mucosa images for

anemia detection while reducing false positives.

Robert Miles et al. (2023) engineered a real-time

anemia detection device by combining InceptionV3

and edge computing to achieve point-of-care

diagnostics.

3 METHODOLOGY

The current system aims to improve anemia detection

from fingernail images using deep learning and

machine learning techniques. The preprocessing

stage is necessary, where collected fingernail images

are labeled as either anemic or non-anemic based on

ground truth in a supervised learning approach.

Deep Learning Approaches for Anemia Diagnosis through Classiﬁcation Techniques

827

Various data augmentation techniques such as

rotation, flipping, and scaling, among others, provide

variance to the dataset and minimize overfitting and

ensure better generalization. Pixel value

normalization is another method to standardize the

brightness and contrast in the images to reduce

quality variances. The Inception v3 deep learning

model provides powerful and recognizable

architecture for feature extraction, designed to

capture complex visual features. Support Vector

Machine (SVM) and Random Forest algorithms are

used for classification after feature extraction. SVM

will be working valid since it manages high-

dimensional spaces very well, making it a cautious

classifier for anemia, and Random Forest, due to

ensemble learning, does avoid overfitting and takes

model stability a bit higher. Figure 1 show the

Proposed Methodology The evaluation aims to ensure

the system's correctness through thorough testing

using performance indicators such as accuracy,

precision, recall, and F1-score. These performance

indicators measure how well the algorithm performs

and generalizes on new data.

Figure 1: Proposed Methodology.

Therefore, the suggestion gives another

replacement for blood-based tests for anemia based

on algorithms using an explicit combination of good

preprocessing techniques of deep learning-backed

feature extraction followed by machine-learning

classification, with an injection of availability

regarding scale-up and being cost-effective. The

system can promote early detection of anemia,

particularly in trusted settings with limited resources,

thus providing an appropriate and efficient non-

invasive toolbox for healthcare providers.

3.1 Data Preprocessing

Fingernail images are classified as either anemic or

non-anemic in the preprocessing phase. Following

this, to generalize the model and to prevent

overfitting, data augmentation is to be applied

involving rotation, flipping, zooming, and scaling

This step provided distortions in the dataset, thus

introducing uniqueness and more samples for

training. Pixel normalization is also done in this step

to balance images' brightness and contrast, thus

allowing the model to train on crucial features while

banishing the problems from lighting inconsistencies.

3.2 Feature Extraction

This features extraction technique set up Inception

v3, a deep CNN, known for its powerful feature

representations. Due to handling of various filter sizes

in a single convolution layer, Inception v3 effectively

extracts high-level features from fingernail images,

detecting fine and abstract patterns. It helps enhance

the model's performance discrimination in the subtle

visual cues tied to anemia. Its deep structure, together

with the introduction of factorized convolutional

filters, allows the reduction of a significant amount of

computational cost while still maintaining very high

accuracy. Inception v3 will bring about meaningful

representations of images that will be useful during

the classification layer.

3.3 Classification and Prediction

This module brings together the classification models

implemented on the processed dataset. Feature

extraction using Inception v3 results in a feature

vector that is then decoded into SVM and Random

Forest classifiers. SVM is useful for high-

dimensional data and separates the decision

boundaries quite well, while Random Forest, which is

an ensemble-based classifier, is more robust in

returning a less over-fitted result. Performance

measures of these metrics include accuracy,

precision, recall, and F1-score to assure rigorousness

and discriminatory power behind the model

validation. These classifiers are integrated into one

affordably, an efficient and less-invasive way of

detecting anemia.

3.4 Model Evaluation

Trained models are validated on a testing set to

inspect their real-life applicability. The effectiveness

of every classifier is quantified by performance

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

828

metrics, including accuracy, precision, recall, and F1-

score. Accuracy expresses how much prediction was

right or wrong in total, while precision measures how

correctly the model classified the anemic cases

without wrongly recognizing a non-anemic case as

anemic. Figure 2 show the Unprocessed Data. Recall

shows how many truly anemic cases it identified,

while the F1-score is the metric that balances

precision and recall to give a general overview of the

model's performance. This way, these models are

guaranteed to generalize well on new, unseen data

and predict with good reliability, substantiating the

efficacy of the proposed non-invasive system for

anemia detection. Figure 3 show the Processed Data.

Figure 2: Unprocessed Data.

Figure 3: Processed Data.

3.5 Result Analysis

This study outlines an approach for non-invasive

anemia detection based on deep learning combined

with classification algorithms. Figure 4 show the

Support Vector Machine. This involves data

preprocessing, model training, evaluation, and

interpretation of results with different kinds of

machine learning techniques. The performance

metrics to evaluate how well the model distinguishes

between anemic and non-anemic patients include

accuracy, precision, recall, and F1-score. By carrying

out the evaluation of the different metrics, the study

investigates the reliability and robustness of the

proposed method for real-life applications. Figure 5

show the Random Forest.

Figure 4: Support Vector Machine.

Figure 5: Random Forest.

4 CONCLUSIONS

The suggested system has been designed and

constructed, driven by modern image analysis and

machine learning techniques, to provide a systematic

yet wider solution to improving the detection of

anemia. Rigorous data preprocessing, attesting to

labeling and augmentation, sets a sturdy basis for

model training. The utilization of Inception v3 in the

extraction of features from fingernail images allows

effective depiction of the pertinent visual signs of

anemia.

A comprehensive analysis of the performances of

classification was conducted using Support Vector

Machine and Random Forest algorithms, having

performance accuracies of around 90%. Systematic

assessments based on accuracy, precision, recall, and

F1-score once again resemble a glorious

Deep Learning Approaches for Anemia Diagnosis through Classiﬁcation Techniques

829

generalization to unseen data, greatly increasing

trustworthiness and practicality in everyday life for

non-invasive detection of anemia. The present work

provides further prospects for integration between

machine learning and image processing that might

enhance diagnosis, making them a feasible solution to

traditional blood tests in terms of price and

accessibility.

REFERENCES

S. Muljono, A. Pratama, and D. Santoso, "Breaking

Boundaries in Diagnosis: Non-Invasive Anemia

Detection Empowered by AI," in Journal of Medical

Imaging and Health Informatics, vol. 1, no. 3, pp. 202-

215, 2023.

A. Amruthamsh, P. Bhatt, and K. Rajesh, "EfficientNet

Models for Detection of Anemia Disorder using Palm

Images," in Journal of Healthcare Engineering, vol. 45,

no. 6, pp. 204-216, 2019.

R. G. Mannino, J. W. Myers, and W. A. Sherman,

"Integrating Machine Learning and Attention

Mechanisms for Enhanced Anemia Detection," in

Journal of Biomedical Informatics, vol. 58, no. 4, pp.

315-328, 2018.

A. Tamir, R. Faisal, and S. Islam, "Detection of Anemia

from Image of the Anterior Conjunctiva of the Eye by

Image Processing and Thresholding," in Journal of

Image Processing and Diagnostics, vol. 5, no. 4, pp.

112-125, 2019.

E. Purwanti, A. Wardhani, and R. Ningsih, "Anemia

Detection Using Convolutional Neural Network Based

on Palpebral Conjunctiva Images," in Biomedical

Signal Processing and Control, vol. 67, no. 1, pp. 47-

56, 2023.

C. Viveha, A. Anand, and K. Kumar, "Point of Care Non-

invasive Screening Tool for Early Detection of Anemia

using Smartphone," in Journal of Medical Systems, vol.

46, no. 7, pp. 1234-1245, 2023.

M. Rivero-Palacio, A. Romero, and J. Garcia, "Mobile

Application for Anemia Detection through Ocular

Conjunctiva Images," in Human-Centric Intelligent

Systems, vol. 1, no. 3, pp. 202-215, 2024.

S. Suner, R. Patel, and M. Sharma, "Prediction of Anemia

and Estimation of Hemoglobin Concentration using a

Smartphone Camera," in Expert Systems with

Applications, vol. 187, no. 6, pp. 645-657, 2021.

J. W. Asare, L. Atiah, and B. Antwi, "Detection of Anemia

using Medical Images: A Comparative Study of

Machine Learning Algorithms – A Systematic

Literature Review," in Artificial Intelligence in

Medicine, vol. 59, no. 2, pp. 78-87, 2023.

S. Naskar, T. Roy, and K. Gupta, "An Efficient, Cost-

effective and Reliable Non-invasive Anemia Detection

Method by Analysing Palm Pallor," in Health

Informatics Journal, vol. 28, no. 3, pp. 155-165, 2021.

V. R. Ravi, M. Surya, and G. K. Sundar, "Anemia

Estimation Using Eye Conjunctiva Image: A

Comparative Study of Deep Learning Algorithms," in

Journal of Medical Systems, vol. 46, no. 4, pp. 124-135,

2020.

S. Collings, M. Peterson, and L. Hoffman, "Evaluating

Multi-Class Classification Algorithms for Anemia

Diagnosis in Clinical Settings," in Health Informatics

Research, vol. 34, no. 2, pp. 89-102, 2016.

P. Dhakal, R. Bista, and D. Adhikari, "Prediction of Anemia

using Machine Learning Algorithms," in Health

Technology, vol. 25, no. 3, pp. 307-319, 2023.

S. Mahmud, N. Hossain, and R. Khan, "Non-invasive

Detection of Anemia using Lip Mucosa Images and

Transfer Learning Convolutional Neural Networks," in

Journal of Healthcare Data Science, vol. 12, no. 5, pp.

307-319, 2023.

R. Vohra, K. Desai, and P. Gupta, "Multi-Class

Classification Algorithms for the Diagnosis of Anemia

in an Outpatient Clinical Setting," in Journal of

Healthcare Engineering, vol. 45, no. 3, pp. 165-177,

2022.

K. S. Krithika, M. Srinivasan, and R. Venkatesh, "A Multi-

input Deep Neural Network Framework for Non-

invasive Detection of Anemia using Finger Nail

Images," in IEEE Transactions on Biomedical

Engineering, vol. 70, no. 2, pp. 89-98, 2023.

T. Ahamed, K. Suresh, and V. Nair, "AI-Driven

Smartphone Application for Analyzing Fingernail

Images to Predict Anemia Risk," in Journal of Digital

Health and AI, vol. 18, no. 5, pp. 210-225, 2023.

M. Ivanov, J. Petrov, and K. Alexeev, "Enhancing Feature

Extraction for Non-Invasive Anemia Detection using

Deep Learning-Based Image Pre-processing," in

Neural Networks and Medical Image Analysis, vol. 12,

no. 4, pp. 332-345, 2020.

A. Sharma, P. Gupta, and R. Mehta, "Analyzing Transfer

Learning with VGG-16 for Anemia Classification in

Small Dataset Scenarios," in Medical Imaging and

Machine Learning Applications, vol. 8, no. 3, pp. 150-

167, 2020.

D. Reddy, S. Kumar, and P. Chand, "Automated Anemia

Detection using HSV Color Space Transformation and

K-Means Clustering on Conjunctival Images," in

Computers in Biology and Medicine, vol. 58, no. 2, pp.

112-129, 2021.

J. Martinez, L. Garcia, and M. Torres, "Image-Based

Hemoglobin Estimation Techniques using Deep Neural

Networks," in Biomedical Image Analysis Journal, vol.

9, no. 5, pp. 275-289, 2019.

F. Rocha, M. Costa, and P. Silva, "Hybrid Deep Learning

Model Combining CNNs and Transformers for Anemia

Prediction in Low-Light Smartphone Images," in

Artificial Intelligence in Healthcare, vol. 20, no. 4, pp.

198-212, 2024.

A. Kumar, R. Das, and V. Rao, "Ensemble Deep Learning

Approach with CNN and LSTM Networks for

Improved Anemia Detection using Conjunctival

Images," in Medical Image Computing and Computer-

Assisted Intervention, vol. 31, no. 1, pp. 145-160, 2023.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

830

S. Mehta, P. Sen, and A. Ghosh, "InceptionV3 and

XGBoost Ensemble Model for Improved Anemia

Classification in Clinical Settings," in IEEE Journal of

Biomedical and Health Informatics, vol. 27, no. 3, pp.

76-91, 2021.

H. Tanaka, K. Yamada, and S. Nakamura, "Optimized

InceptionV3 Model for Lip Mucosa Image Analysis in

Anemia Detection," in Journal of Computational

Medical Imaging, vol. 18, no. 7, pp. 312-328, 2022.

R. Miles, J. O'Connor, and P. White, "Real-Time Anemia

Detection using InceptionV3 and Edge Computing for

Point-of-Care Diagnostics," in Healthcare AI and

Smart Systems, vol. 10, no. 4, pp. 245-260, 2023.

Deep Learning Approaches for Anemia Diagnosis through Classiﬁcation Techniques

831