Automated Detection of Diabetic Foot Ulcers and Multi-Level Severity Grading Using an Intelligent Deep Learning Framework

Automated Detection of Diabetic Foot Ulcers and Multi-Level Severity Grading Using an Intelligent Deep Learning Framework

IEEE BASE PAPER TITLE:

An Emerging Lightweight Large-Scale Attention-Attenuated Deep Model for Real-Time Segmentation and Predictive Analytics of Foot Ulcers

IEEE BASE PAPER ABSTRACT:

The integration of consumer electronics with intelligent healthcare systems has opened new possibilities for real-time disease monitoring and remote clinical assessment. Among diabetes-related complications, Diabetic Foot Ulcers (DFUs) remain a major global concern, often leading to infection, hospitalization, or amputation when not identified early. This study presents an Explainable Lightweight Dual-Encoder Attention-Attenuated Framework that bridges advanced deep learning with consumer-grade medical imaging devices for automated DFU segmentation and analysis. The proposed system fuses the global contextual learning of the Swin Transformer encoder with the fine-grained local representation of the ResNeXt encoder, while a Self-Aware Attention (SAA) mechanism enhances feature interpretability and boundary refinement. Furthermore, the inclusion of Explainable AI (XAI) modules allows clinicians to visualize attention responses and understand the model’s diagnostic reasoning, thus improving transparency and trust in automated predictions. Designed for deployment on portable and edge-enabled healthcare electronics, the framework achieves Dice scores of 0.975 and 0.977 on two benchmark DFU datasets while maintaining a compact computational footprint. These results demonstrate the potential of the proposed architecture as an interpretable, resource-efficient, and clinically reliable AI solution for real-time diabetic foot ulcer monitoring within consumer healthcare ecosystems.

PROJECT OUTPUT VIDEO:

ALGORITHM / MODEL USED:

DenseNet121, MobileNet

OUR PROPOSED ABSTRACT:

Diabetic foot ulcers (DFUs) are a serious complication of diabetes that can progressively damage the skin and underlying tissues of the foot. Early identification and assessment of ulcer severity are important for supporting timely intervention and reducing the risk of further complications. However, manual examination of foot images can be time-consuming and may be influenced by variations in image quality and clinical interpretation. To address this need, this project presents an Automated Detection of Diabetic Foot Ulcers and Multi-Level Severity Grading Using an Intelligent Deep Learning Framework, which utilizes deep learning-based medical image classification to automatically analyze foot images and determine their corresponding condition and severity level.

The dataset used in the project contain foot images with five-class problem consisting of Grade 0, Grade 1, Grade 2, Grade 3, and Normal (Healthy skin). The dataset contains 1,081 images, comprising 965 training images and 116 validation images, with an additional false-case set containing 207 images of abrasions, bruises, burns, and cuts.

The proposed system employs two deep learning architectures, MobileNet and DenseNet121, to perform automated image classification. The final evaluation shows that MobileNet achieves 94.8% accuracy, whereas DenseNet121 achieves 97.4% accuracy, making DenseNet121 the better-performing model in the reported evaluation.

To provide a practical interface for image-based prediction, the trained models are integrated into a Flask web application developed using Python, with HTML, CSS, and JavaScript used for the front end and SQLite used as the database. The model generates class probabilities, and the class with the highest probability is selected as the predicted result along with its confidence percentage.

The developed framework therefore provides an automated approach for identifying diabetic foot ulcer conditions and performing multi-level severity grading from Grade 0 to Grade 3, while also distinguishing normal healthy skin. By incorporating and comparing MobileNet and DenseNet121 within a Flask-based web application, the system combines deep learning-based image analysis with an accessible web interface for prediction and performance assessment. The reported results demonstrate the effectiveness of the proposed approach, with DenseNet121 achieving the highest classification accuracy of 97.4% among the two evaluated models.

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS: 

• System : Pentium i3 Processor.
• Hard Disk : 20 GB.
• Monitor : 15’’ LED.
• Input Devices : Keyboard, Mouse.
• Ram : 8 GB.

SOFTWARE REQUIREMENTS: 

• Operating System : Windows 10 / 11.
• Coding Language : Python 3.13.15.
• Web Framework : Flask.
• Frontend : HTML, CSS, JavaScript.
• Database : SQLite.

REFERENCE:

Nishu Bansal and Ankit Vidyarthi, “An Emerging Lightweight Large-Scale Attention-Attenuated Deep Model for Real-Time Segmentation and Predictive Analytics of Foot Ulcers”, IEEE TRANSACTIONS ON CONSUMER ELECTRONICS, VOL. 72, NO. 1, FEBRUARY 2026.

👉CLICK HERE TO BUY THIS PROJECT “Automated Detection of Diabetic Foot Ulcers and Multi-Level Severity Grading Using an Intelligent Deep Learning Framework” SOURCE CODE👈

Frequently Asked Questions & Answers:

What is the main objective of this project?

The main objective of the project is to develop an automated deep learning-based system for analyzing diabetic foot images and classifying them according to their condition and severity. The system performs five-class classification into Grade 0, Grade 1, Grade 2, Grade 3, and Normal (Healthy Skin).

What is a Diabetic Foot Ulcer?

A diabetic foot ulcer is a foot wound or lesion associated with diabetes. The project focuses on analyzing foot images to identify the classification category represented by the image.

How many images are used in the project?

The dataset contains 1,081 images, including 965 training images and 116 validation images. The dataset also contains an additional 207 false-case images representing abrasions, bruises, burns, and cuts.

Why are MobileNet and DenseNet121 used?

The project uses these two architectures to perform diabetic foot image classification and compare their classification performance. Both models are trained independently and integrated into the web application for prediction.

Which model achieved the highest accuracy?

DenseNet121 achieved the highest reported accuracy of 97.4%, while MobileNet achieved 94.8%.

How does the prediction process work?

The user uploads a foot image and selects either MobileNet or DenseNet121. The application validates and saves the image, resizes and normalizes it, and sends it to the selected model. The model generates probabilities for the five classes, and the class with the highest probability is selected as the prediction.

How is the proposed system different from the earlier system?

The earlier system described in the base paper focuses on continuous diabetic foot monitoring using thermal and pressure measurements collected through a smart insole and analyzes synchronized spatiotemporal data using a hybrid CNN-Transformer model. The proposed project instead focuses on image-based deep learning classification using MobileNet and DenseNet121, with five classification categories covering Normal and Grade 0 to Grade 3

Can this project be used as an academic final-year project?

Yes. The project combines Artificial Intelligence, Deep Learning, Medical Image Processing, Python, Flask, HTML, CSS, JavaScript, and SQLite, making it suitable as an academic project involving machine learning and web application development.

Can a research paper be developed from this project?

Yes. The project contains a defined research problem, dataset, deep learning methodology, comparative model evaluation, and measurable results. The reported performance of MobileNet and DenseNet121 can form part of an academic research study, subject to appropriate validation, documentation, and research methodology. JP INFOTECH provides assistance in preparing International Journal Paper, International Conference Paper or IEEE Conference Paper.