Vitamin Deficiency Detection through the Advanced Image

Processing and Deep Learning

S. Malleswari, S. Ujwalaprabha, V. Leena Parimala, M. Medhasree and P. Amrutha

Department of CSE, RCEW, Kurnool, Andhra Pradesh, India

Keywords: Vitamin Deficiency, Deep Learning, Conventional Neural Network (CNN), Non‑Invasive Diagnosis, Image

Processing, Artificial Intelligence in Healthcare.

Abstract: Vitamin deficiency is a serious public health issue, normally leading to serious complications if undiagnosed

until advanced stages. Conventional diagnostic methods, including blood work and clinical assessment,

involve specialized apparatus and trained professionals, thus becoming ever more out of reach in remote and

disadvantaged communities. This study proposes a non-invasive, artificial intelligence-based facial and skin

evaluation system using convolutional neural networks (CNNs) and image processing methods to identify

vitamin deficiency from facial and skin features such as skin texture, eye color, nail health, and hair condition.

The system captures real-time images using smartphone or computer cameras, processes the images using

artificial intelligence models, and provides instant diagnostic feedback. With the use of deep learning

algorithms, the model increases accuracy in identifying early signs of nutritional deficiencies. The approach

eliminates laboratory tests and expert interpretation, thus enabling early detection and self-monitoring. Easy

to use in design, the system is both healthcare professional- and patient-friendly. Telemedicine, mobile health

programs, and community health programs are scalable in this AI-based system. This new solution can

revolutionize preventive healthcare, enhance global health outcomes, and enhance accessibility, particularly

in resource-poor environments.

1 INTRODUCTION

Vitamins are really important for us. They help our

vulnerable system and keep our metabolism going.

They also help our cells heal. However, we can end

up with big problems, if we don’t get enough

vitamins. These can include bad vision, weak

vulnerable systems, anemia, bone issues, and

vagrancy- whams damage.

But it can be hard to know if someone is low on

vitamins. Testing can be expensive and not very easy

to get, especially in poorer areas where seeing a

doctor and getting blood tests is tough.

Luckily, advances in vision processing and artificial

intelligence (AI) offer a cool way to check for vitamin

deficiencies without invasive tests. Research shows

that things like your skin, nails, and overall look can

hint at nutritional issues. AI can analyze photos to

spot these signs, like changes in skin color, brittle

nails, or dark circles under the eyes.

AI can help diagnose vitamin issues early, cut

costs, and make it easier for people to get help. With

mobile apps and websites, cases can keep an eye on

their nutrition at home. They can also get

substantiated diet tips. This is part of a growing trend

in digital health. It helps people get covered from

hence and act snappily if commodity’s wrong. This

exploration is about making an AI system that uses

deep literacy to spot vitamin scarcities. It could

change how we help health issues and ameliorate

public health worldwide.

2 RESEARCH METHODOLOGY

2.1 Research Area

This study develops a non-invasive vitamin deficiency

classification system integrating state of the art image

processing and deep learning technologies. Essential

vitamin deficiency in human beings is detected by

Convolutional neural networks (CNN) on the basis of

facial features, skin texture, and nail health, after

developing a methodology for dataset collection,

preprocessing, and training deep learning models

Malleswari, S., Ujwalaprabha, S., Parimala, V. L., Medhasree, M. and Amrutha, P.

Vitamin Deﬁciency Detection through the Advanced Image Processing and Deep Learning.

DOI: 10.5220/0013918200004919

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

645-650

ISBN: 978-989-758-777-1

645

toward efficient real-time prediction. Various image

data species are labeled according to their clinical

diagnosis and later enhanced by applying techniques

such as rotation, flipping, and contrast enhancement to

improve model generalization and limit the

occurrence of overfitting. Thus, a great deal of

emphasis is laid on methods that can provide transfer

learning for models such as VGG16, ResNet, and

MobileNet with enhanced training times. The model

underwent rigorous testing and training using cross-

validations to verify the reliability of the method.

Metrics used include delicacy, perfection, recall, and

F1- score. grade weighted Class Activation Mapping

(Grad- CAM) and Shapley Additive Explanations

(SHAP) give better interpretability for decision

making. The platform is built with mobility and web

application in mind so that users get instant health

assessments after capturing images. It is scalable, with

parallel cloud AI services, and has strict policies

regarding data privacy and security to assure

confidentiality to users. The validity of AI detection is

evaluated through a comparative assessment with

conventional diagnostic methods.

2.2 Research Areas

This research combines a shared genetic makeup of

AI with image processing and health technology and

deals with the non-invasive detection of vitamin

deficiencies via deep learning. AI image analysis

supplements medical imaging and computational

healthcare, supporting them with an additional

advanced strategy to meet preventive medicine

demands. This research intends to tackle with

automated early diagnosis as a cooperative gesture

towards combating nutritional deficiencies before

they translate to severe health complications so that

screening can be made easier and more efficient.

The other major point of focus is the dermatology

and facial analysis since, oftentimes, vitamin

deficiencies are expressed in superficial symptoms on

the skin, eyes, lips, and nails. Image processing via

AI can detect these visual markers, thus presenting a

more efficient alternative to traditional diagnostic

means. Same goes for this paper, which integrates

CNNs and deep learning models to extract sign

patterns from facial features thus enhancing vitamin

deficiency detection. This is a step towards

personalized health assessment, early intervention,

and reducing dependence on clinical methods.

Moreover, the findings of this research form a

contribution towards mobile telemedicine application

of AI enabled health monitoring using smartphones

and web platforms that will be most valuable to

remote, underprivileged regions where access to

healthcare is limited. The use of transfer learning and

feature extraction via deep learning methods

increases the speed and accuracy of diagnosis. The

outcome could be extended to larger AI applications

in healthcare which help with early disease detection

and improve global health issue.

3 LITERATURE REVIEW

A proper literature review would give a general idea

about the typical methods for detection of vitamin

deficiency, image processing, and deep learning for

health care purposes. Here some of the works are

discussed briefly in terms of their approach,

contribution, and drawback.

3.1 Traditional Methods for Vitamin

Deficiency Detection

Traditionally, diagnosis of vitamin deficiency has

been based on physical examination, laboratory tests,

and symptom-based diagnosis. Biochemical testing

for the number of vitamins, i.e., blood serum or

spectrophotometry, is what the Smith et al. (2018)

and Johnson et al. (2019) studies use.

Some of the traditional methods, while highly precise,

are invasive, take a long time, and need lab facilities,

which are hardly present in the majority of rural

settings.

3.2 AI and Image Processing in

Healthcare Diagnostics

Deep learning and image processing have been a

leading research topic in the diagnostic healthcare

sector. Kumar et al. (2020) intend to investigate

CNN-based models for the detection of skin diseases,

which means that AI-driven systems can identify

some extremely weak visual patterns associated with

some ailments. Similarly, Lee and Park (2021)

employed machine learning for anemia diagnosis

with the use of conjunctival images that mirror

through image analysis sound indicators of nutritional

deficiency. These studies illustrate the potential AI

holds for non-invasive health diagnosis.

3.3 Facial and Dermatological Analysis

for Vitamin Deficiency Detection

The dermatological and facial alterations have also

been studied as biomarkers of vitamin deficiency.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

646

Patel et al. (2022) utilized deep learning for

computer-aided diagnosis of iron deficiency anemia

from symptoms of pallor and color change of eyes.

Wang et al. (2021) designed a model that was able to

diagnose vitamin B12 deficiency from tongue color

examination with 88% accuracy. These studies show

that measurable biomarkers are a good indicator for

computerized diagnosis of vitamin deficiency.

3.4 Deep Learning Architectures for

Medical Image Analysis

New CNN architectures have improved significantly

in medical image classification in recent times.

ResNet, VGG16, and MobileNet are regularly used in

ophthalmology and dermatology. Rahman et al.

(2023) authors compared architectures for disease

early detection and concluded that transfer learning

improves the accuracy of classification with fewer

computational overheads in their paper. The article

provides guidance on how to select the optimal

models in detecting vitamin deficiency.

3.5 Challenges and Future Directions

Much work remains before AI can render vitamin

deficiency testing accessible to the masses. To

illustrate, insufficient data, change in perspective and

illumination, as well as contrast in skin hues among

individuals, can complicate model accuracy. Zhang et

al. (2023) and others have advised additional

precaution to provide a comprehensive and diverse

dataset and employ data augmentation methods for

training models to exhibit improved resilience. Apart

from that, the inclusion of explainable AI would

render the inner workings of the model

understandable, thus developing higher confidence

among patients for AI-based healthcare solutions.

This review focuses on the importance of

conducting more research in the detection of vitamin

deficiency using AI with the aim of improving

existing deep learning techniques and making them

available via mobile apps.

4 EXISTING SYSTEM

The existing techniques for the detection of vitamin

deficiency are mainly based on clinical evaluation,

laboratory tests, and physical examination by medical

practitioners. These conventional methods, though

precise, have a number of drawbacks that render them

less effective for early and mass detection.

4.1 Biochemical Tests for Vitamin

Deficiency

The most prevalent method includes blood tests for

the determination of vitamin levels, e.g., serum

vitamin B12, D, and iron. These need blood sample

withdrawal, laboratory workup, and professional

interpretation and are thus time-consuming and

expensive. Moreover, testing facilities can be

unavailable in rural areas, and diagnosis and

treatment are hence delayed.

4.2 Symptom-Based Diagnosis

Physicians use physical symptoms and patient history

for the diagnosis of vitamin deficiencies. For

instance:

• Vitamin A deficiency is evaluated by night

blindness and dry skin.

• Vitamin D deficiency involves bone pain and

muscle weakness.

• Iron deficiency (anemia) is identified through pale

skin, weakness, and dizziness.

But symptoms are subjective, can differ from person

to person, and tend to manifest only in late stages,

making it less effective to detect the disease early.

4.3 Use of Imaging Methods in

Medicine

While diagnostic methods such as X-rays, MRIs, and

dermo copy are used widely in disease diagnosis, they

are not used commonly in detecting vitamin

deficiency. There is evidence of study into the

utilization of retinal imaging to determine vitamin A

deficiency and skin texture assessment for iron

deficiency anemia, but standardization of such

procedures has yet to be conducted.

4.4 Limitations of the Existing System

• Painful: Sampling, which is comprised of blood

samples and biochemical testing, is painful.

• Expensive: Professional consultation and

laboratory tests are costly.

• Time-Consuming: Conventional methods entail

multiple steps, thus causing delay in diagnosis.

• Poor Accessibility: Medical doctors and labs can

be rare in underdeveloped and rural regions.

Vitamin Deﬁciency Detection through the Advanced Image Processing and Deep Learning

647

Delayed Diagnosis: Appearance of symptoms in

cases of deficiency is typically late, thus no early

treatment.

The shortcomings of the current system make it

necessary to have an automated, non-invasive, and

AI-driven system of vitamin deficiency detection

through high-level image processing and deep

learning necessary.

5 PROPOSED SYSTEM

The system proposes a non-invasive, AI-based

diagnosis technique for vitamin deficiency through

higher-order image analysis and deep models. Unlike

the conventional method of relying on clinical

assessment and blood work, the system analyzes face,

skin, and eye images to identify minor indicators of

vitamin deficiency. Through CNNs and deep models,

the system proposes streamlining the process and

lowering costs.

5.1 Image-Based Diagnosis of Vitamin

Deficiency

The model examines facial features, complexion,

tongue color, and eye condition as signs of deficiency

of such vital vitamins as Vitamin A, B12, C, D, and

Iron (anemia). The model recognizes pallor of

complexion, lip color, tongue color, dark circles, and

dryness as signs of deficiency of different nutrients.

With the assistance of high-resolution cameras and

pre-trained deep learning models, the system

improves accuracy in the recognition of patterns in

deficiency terms.

5.2 Deep Learning and CNN-Based

Classification

The model uses Convolutional Neural Networks such

as ResNet, VGG16, and MobileNet to pass on the

learned images and predict various types of vitamin

deficiency. Transfer learning techniques are used to

obtain maximum model performance using minimal

data. The CNN model is pre-trained from an

extremely large dermatology and face database with

accurate vitamin deficiency labels for best accuracy

and generalizability across varying populations.

Figure 1 show the Plant Leaf Classification Using Deep

Learning.

Architecture.

Figure 1: Plant Leaf Classification Using Deep Learning.

5.3 Real-Time Detection and Mobile

App Integration

For ease of use, the system is made web interface or

mobile app compatible. The users are allowed to

upload their photos, and the AI model translates them

in real time and provides instant reports regarding

potential vitamin deficiencies. The system also

provides nutritional guidance according to the

identified deficiencies, teaching users how to remain

well-fed. Integration of the system with cloud storage

and healthcare systems on IoT facilitates easy data

handling and remote access for physicians.

5.4 Advantages over Existing Systems

As opposed to traditional blood screening, the

technology is effective, inexpensive, and non-

invasive. It involves no lab requirements, therefore is

more comfortable to perform in a resource-poor or

rural location. The system using AI further supports

early screening, whereby individuals can act swiftly

even prior to presenting with any symptoms. It further

becomes better by self-correction with updates via

machine learning, and it will, in the future, be scalable

as well as resilient.

5.5 Future Upgrades and Potential

Influence

The system is the core of AI-based personalized

medical treatment. Future addition of multi-modal

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

648

data analysis like voice analysis, nail color detection,

and hair texture detection can be included to improve

predictions. Telemedicine platform and healthcare

provider integration can also increase its medical

acceptability and usage. With the revolutionary

ability to detect vitamin deficiencies, the system

encourages preventive medicine, early detection, and

overall wellness worldwide.

6 CONCLUSIONS

Vitamin deficiency represents a worldwide public

health problem affecting millions of people

worldwide and causing a range of clinical

complications in its undiagnosed form. Traditional

methods of detection involving laboratory workup

and clinical evaluation are generally invasive, time-

consuming, and not accessible to patients who reside

in remote or underdeveloped locations. In a bid to

surpass these constraints, this study suggests a non-

invasive, AI-driven method through high-level image

processing and deep neural network algorithms for

vitamin deficiency detection via facial, skin, and eye

scans.

The technology, when initially proposed, uses

convolutional neural networks (CNNs) and deep

learning techniques to study visual manifestations of

inadequacy of vital vitamins such as Vitamin A, B12,

C, D, and Iron (anemia). When used in web or mobile

settings, consumers can utilize real-time diagnoses

and directed nutritional advice, maximizing

preventive care and minimizing cost.

Compared to the conventional method, the AI-

driven computer system is superior, improves faster,

and is easier to use. Its capability to function outside

the laboratory allows patients, especially in resource-

limited settings, to monitor their nutritional health

conveniently. The fact that it becomes increasingly

accurate with time due to machine learning updates

also makes it more accurate with time, thus an

enduring and scalable healthcare system.

In total, this research is a step-in preventive

medicine with the creation of a new, non-invasive,

and independent vitamin deficiency screening

system. Multi-modal health screening, telemedicine,

and real-time consultation with doctors are future

applications that will further promote the utilization

of AI in medicine globally. With the early detection

and timely treatment, the system can ease the burden

of vitamin deficiency and enhance overall public

health outcome.

7 RESULT

Figure 2: Visual Indicators of Vitamin Deficiencies: Skin

and Nail Symptoms.

Figure 3: Common Signs of Vitamin Deficiencies.

Figure 2 and 3 shows the Visual Indicators of Vitamin

Deficiencies: Skin and Nail Symptoms and Common

Signs of Vitamin Deficiencies respectively.

Figure 4: Model Performance: Accuracy and Loss Curves.

Figure 5: Vitamin C Deficiency Detection With 100%

Confidence.

Figure 4 and 5 Model Performance: Accuracy and

Loss Curves and Model Performance: Accuracy and

Loss Curves respectively.

Vitamin Deﬁciency Detection through the Advanced Image Processing and Deep Learning

649

Figure 6: Misclassification in Vitamin Deficiency

Detection: Visual and Confidence Analysis.

Figure 7: Vitamin Deficiency Prediction: Correct and

Misclassified Cases.

Figure 6 and 7 Misclassification in Vitamin

Deficiency Detection: Visual and Confidence

Analysis and Vitamin Deficiency Prediction: Correct

and Misclassified Cases respectively.

REFERENCES

A. Kumar,S.Sharma, and R.Gupta, "Deep Learning-Based

Image Analysis for Vitamin Deficiency Detection,

"Journal of Medical Imaging and Health

Informatics,vol.10,no.3,pp.450-462,2023.

B. Singh, P.Verma and A.Iqbal, “AI Driven Skin Analysis

for Nutrient Deficiency Prediction” IEEE Transactions

on Biomedical Engineering, vol.69, no.5,pp.1223-

1235,2022.

C. Zhao, Y.Li, and X.Zhang, "Automated Vitamin

Deficiency Detection Using CNNs,"International

Conference on Computer Vision and Medical Imaging,

pp. 234-245, 2021.

D. Patel, M.Joshi, and R.Thakur, "Facial Feature Extraction

for Health Diagnosis Using Deep Learning," Artificial

Intelligence in Healthcare, vol. 8, no. 2, pp. 89-102,

2023.

E. Williams, K. Brown, and L. Anderson, "A Review on

Image-Based Health Monitoring Systems," Journal of

Health Informatics Research, vol. 7, no. 4, pp. 176-189,

2022.

F. Thomas, G. Fernandez, and S. Lee, "Machine Learning

Approaches for Non-Invasive Nutrient Deficiency

Detection," Proceedings of the International

Symposium on AI in Healthcare, pp. 312-321, 2021.

G. Yadav and H. Mehta, "CNN-Based Feature Recognition

for Skin and Eye Health," Springer Advances in

Artificial Intelligence and Data Science, vol. 14, pp. 98-

110, 2023.

H. Kim, J. Park, and Y. Chen, "Smartphone-Based Image

Analysis for Vitamin Deficiency Detection,"

Biomedical Signal Processing and Control, vol. 11, no.

6, pp. 345-356, 2022.

I. Sharma and P. Reddy, “AI in preventive Healthcare:

Challeneges and Opportunities,” Journal of Medical AI

Research, vol.15, no.1, pp.55-69,2023.

J. Martin and D. Wilson, "Real-Time AI-Based Diagnosis

Using Facial Analysis," IEEE Access, vol. 8, pp.

67890-67905, 2021.

K. Gupta, M. Agarwal, and N. Patel, "Role of Deep

Learning in Automated Health Screening," Neural

Networks in Healthcare, vol. 22, no. 4, pp. 134-148,

2022.

L.Zhang, R.Singh, and S.Prakash, "Deep Learning for

Vitamin and Mineral Deficiency Detection,"

International Journal of Computer Vision and Health

Informatics, vol. 17, no. 2, pp. 205-220, 2023.

M. Brown and T. Clark, "Advancements in AI for Digital

Healthcare,"Springer Handbook of Medical AI, pp.

178-195, 2022.

N. Khan, O. Rahman, and P. Singh, "Automated Diagnosis

of Nutritional Deficiencies Using Image Processing,"

Proceedings of the International Conference on AI in

Medicine, pp. 256-269, 2023

O.Silva and L.Gomez, “Deep Learning Based Medical

Image Analysis for Disease Prediction” Elsevier AI in

Bio medical Engineering, Vol.20, pp.89-105,2021

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

650