INTEGRATION OF DEEP LEARNING AND

IOT FOR REAL TIME BRAIN TUMOR

DETECTION
Content
• Introduction
• Points to know
• Literature Survey
• Problem Statement
• Objectives
• Methodology
• Challenges and Solutions
• Conclusion
• References

2 Department of EEE 05/28/202

5
Introduction

• Brain tumors are serious health conditions that can be life-threatening because
they affect important brain functions. Detecting them early and accurately is
essential for better treatment, but traditional methods often face challenges, such
as slow results, reliance on skilled doctors, and limited access in remote areas.
• To solve these problems, this project combines Deep Learning (DL) and the
Internet of Things (IoT) to create a real-time brain tumor detection system. By
using IoT devices to collect data and a CNN-based Deep Learning model to
analyze the information, the system automatically detects tumor cells in MRI
scans with high accuracy.
• This new approach not only provides faster and more accurate results but also
makes healthcare more accessible in remote areas. With the power of Deep
Learning for recognizing patterns and IoT for easy data transfer, this project aims
to improve medical diagnostics and offer a cost-effective way to provide better
healthcare for everyone.

3 Department of EEE 05/28/2025

 Literature Survey
Sl. No. Details of the Paper Methodology Outcome

Koshti, S., et al. Brain Tumor Used CNN with transfer

learning to classify MRI Achieved 97% accuracy with
Detection System using
1 scans into tumor categories DenseNet201. Web app enables efficient
Convolutional Neural Network. Developed a web app for tumor detection in hospitals.
2022.. deployment.

Utilized a VGG-16 CNN

model trained on
Siddique, M. A. B., et al. Deep
augmented MRI data. Achieved 96% accuracy, with an F1-
Convolutional Neural Networks
2 Preprocessing included score of 0.97, and AUC of 0.95. Suitable
Model-based Brain Tumor Detection
normalization and for clinical use.
in Brain MRI Images. 2020.
thresholding to remove
noise.

various methods like

thresholding region
Anurag goswami & manish dixit , Improved detection accuracy and
based, edge based , fuzzy
Analysis of image segmentation comprehensive comparison of method
3 theory , ANN were
techniques for brain tumor for Brain Tumor segmentation and
explored to enhance
detection,2020 classification
accuracy and processing
speed

4 Department of EEE 05/28/202

5
 Continue…

Sl. No.. Details of the Paper Methodology Outcome

Proposed an anomaly detection

Sudha Varalakshmi et al.,
model for IoT networks using CNN Achieved 98.5% accuracy with
"Design of IoT Network Using
(1D, 2D, 3D). Transfer learning and improved F1 score, precision, and
4 Deep Learning-Based Model
model tuning optimized recall for multiclass and binary
for Anomaly Detection," IEEE
performance. Validated with traffic classification.
I-SMAC
datasets like IoT-DS, BoT-IoT,

5 Department of EEE 05/28/202

5
 Problem Statement
 What’s the Issue?
• Brain tumors are life-threatening conditions, and their timely detection is crucial for
effective treatment. However, traditional diagnostic methods rely heavily on manual
interpretation of MRI scans, which is time-consuming, prone to human error, and often
inaccessible in remote areas.
 Challenges:
• Delayed diagnosis due to a lack of automated systems.
• High dependency on skilled radiologists, which limits access in underserved regions.
• Difficulty in processing large volumes of MRI data quickly and accurately. The absence of
real-time, scalable solutions for early detection.
 Problem to Solve:
o This project aims to address these challenges by creating an automated brain tumor
detection system using Deep Learning and IoT. It provides:
• Real-time data analysis for faster diagnosis.
• High accuracy in tumor classification using a CNN-based model.
• Remote accessibility through IoT, ensuring even remote areas benefit from advanced
diagnostics.
• A scalable and cost-effective solution for improving healthcare outcomes.
6 Department of EEE 05/28/2025
Objectives
• Primary Goals
• Develop an automated system for brain tumor detection and
classification
• Create a user-friendly web interface for medical professionals

• Technical Objectives
• Implement a hybrid machine learning model
• Achieve high accuracy in tumor classification
• Ensure efficient processing of MRI images

• Clinical Objectives
• Provide a reliable diagnostic support tool
• Reduce time-to-diagnosis
• Improve accessibility of specialized analysis

7 Department of EEE 05/28/2025

DATASET SAMPLE
🞂 MRI SCANS:

8 Department of EEE 05/28/2025

Methodology - Data Collection

• Data Sources
• Public medical imaging repositories
• Hospital archives
• Four-class classification:
1.Glioma (a type of brain tumor).
2.Meningioma (another type of brain tumor).
3.Pituitary tumor (affects the pituitary gland).
4.No tumor (healthy scans).
• Preprocessing Steps
• Noise reduction (Gaussian filtering): Blurs unnecessary
details
• Skull stripping: Removing the skull from MRI images so the
model focuses only on brain tissue.
• Data augmentation: Increases the dataset size by creating
modified versions of existing images

9 Department of EEE 05/28/2025

Methodology - Feature Extraction
• Handcrafted Features
• Texture analysis (GLCM): Gray Level Co-occurrence Matrix,
Measures patterns and variations in the tumor's texture.
• Shape features: Extracts tumor size, edges, and irregularities.

• Automatic Feature Extraction

• Pretrained CNN models ( ResNet50): Uses an existing deep
learning model trained on large image datasets to recognize
patterns.
• Transfer learning techniques: The model is fine-tuned
using brain MRI images, adapting it to the new task.

10 Department of EEE 05/28/2025

Methodology - Model Development
Deep Learning Pipeline
• Dimensionality reduction: Reducing unnecessary image
details to speed up training.
• Traditional classifiers (SVM, Random Forests):Machine
learning models used alongside deep learning for better
performance.
• Hybrid models (Fuzzy-ANN):A combination of fuzzy logic
and artificial neural networks to handle uncertainty in tumor
classification.
• Deep Learning Architecture
• Custom CNN with 12 convolutional layers
• 3D CNN for volumetric analysis.
• Transfer learning integration

11 Department of EEE 05/28/2025

Methodology - Implementation
• Web Application (Django Framework)
• Backend: API endpoints, model serving, database
management
• Frontend: Responsive design, visualization components, user
authentication
• Model Deployment
• Cloud as IOT.
• Swagger UI for API documentation,

12 Department of EEE 05/28/2025

Methodology - Evaluation

• Performance Metrics
• Accuracy
• Confusion matrix
• ROC-AUC
• Dice Similarity Coefficient
• Clinical Validation
• Radiologist review
• Cross-institutional testing

13 Department of EEE 05/28/2025

Expected Result:
Brain Tumor Detection and
Classification

15 Department of EEE 05/28/2025

Use case with architecture

16 Department of EEE 05/28/2025

Work Done So Far
• Web Application Implementation
• Set up Django project structure
• Created image upload functionality
• Integrated DL model with web backend
• Developed basic user interface for predictions.

• Training & Initial Results

• Conducted initial training sessions
• Achieved 60-75% accuracy on test set
• Identified areas for model improvement
• Implemented basic performance monitoring

17 Department of EEE 05/28/2025

Challenges and Solutions

• Class Imbalance
• Data augmentation + weighted loss functions.

• Computational Complexity
• Model quantization and pruning.

• Interpretability
• Explainable AI (XAI) techniques (Grad-CAM)

18 Department of EEE 05/28/2025

Model accuracy

19 Department of EEE 05/28/2025

Model Loss

20 Department of EEE 05/28/2025

Test and train split

21 Department of EEE 05/28/2025

Image classes

22 Department of EEE 05/28/2025

Present output

23 Department of EEE 05/28/2025

Future Work
Model Improvement
• Hyperparameter tuning
• Advanced architectures (ResNet)

Web Application Enhancements

• User authentication
• Detailed visualization tools
• Performance optimization

Clinical Validation
• Radiologist review
• Multi-institutional testing

24 Department of EEE 05/28/2025

References
1. 1. Mohammadi M et al (2018) Deep learning for IoT big data and streaming analytics: a
survey.
2. Kapoor L, Thakur S (2017) A survey on brain tumor detection using image processing
techniques.
3. Kothari A, Indira B (2015) A study on classification and detection of brain tumor techniques.
4. Zacharaki E, Wang S, Chawla S, Soo Yoo D, Wolf R, Melhem E, Davatzikos C (2009)
Classifica tion of brain tumor type and grade using MRI texture and shape in a machine
learning scheme.
5. Abd-Ellah M, Awad AI, Khalaf AAM, Hamed H (2020) Deep convolutional neural networks.
6. Hemanth DJ, Anitha J, Naaji A, Geman O, Popescu DE (2018) A modified deep
convolutional neural network for abnormal brain image classification.
7. Çinar A, Yildirim M (2020) Detection of tumors on brain MRI images using the hybrid
convolu tional neural network architecture.
8. Tazin T, Sarker S, Gupta P (2021) A robust and novel approach for brain tumor classification
using convolutional neural network.
9. Ilunga-Mbuyamba E, Avina-Cervantes JG, Cepeda-Negrete J, Ibarra-Manzano MA,
Chalopin C (2017) Automatic selection of localized region-based active contour models
using image con tent analysis applied to brain tumor segmentation.
10. El-Feshawy S, Saad W, Shokair M, Dessouky M (2022) Proposed Approaches for Brain
Tumors Detection Techniques Using Convolutional Neural Networks.

25 Department of EEE 05/28/202

5
 Thank You

26 Department of EEE 05/28/2025