IEEE SCEECS 2026 · PEER-REVIEWED RESEARCH PROTOTYPE

DETECT
BRAIN
TUMORS

Upload any brain MRI scan. Get AI-powered classification across four clinical categories — Glioma, Meningioma, Pituitary, and No Tumor — backed by a published IEEE paper achieving 99% accuracy.

⚠ RESEARCH PROTOTYPE · NOT FOR CLINICAL USE
AUC 0.999
NEURAL NET
16K+ SCANS
99.0% ACC
5cm L R S
0 % Classification Accuracy
10-Fold Cross-Validation
0.999 AUC Score
Neural Network
0 + Training Images
Multi-Source Dataset
0 ML Models
Benchmarked
How It Works

THREE STEPS TO
CLASSIFICATION

01
UPLOAD MRI
Drag and drop your brain MRI scan — T1, T2, or any axial/coronal/sagittal view. Any resolution. The system handles all preprocessing internally: 224×224 resize, pixel normalization, contrast enhancement.
02
AI ANALYSIS
Claude Vision AI performs deep learning analysis using InceptionV3 feature embeddings — the same architecture used in the IEEE paper. Features are projected through PCA reduction and classified by the neural network.
03
FULL REPORT
Receive the detected tumor type, per-class confidence scores, clinical risk level, and a plain-English explanation of the findings. Download a print-ready PDF diagnostic report for further consultation.
Detection Categories

FOUR CLINICAL
CATEGORIES

🔴
GLIOMA
Tumors arising from glial cells. The most common primary brain tumor, often aggressive. Includes Glioblastoma Multiforme (GBM) — grade IV. Requires immediate specialist referral.
HIGH RISK
🟡
MENINGIOMA
Arises from the meninges surrounding the brain and spinal cord. Usually benign and slow-growing. Most common primary brain tumor in adults. Often surgically treatable with excellent prognosis.
MEDIUM RISK
🟣
PITUITARY
Tumor on the pituitary gland at the brain base. Disrupts hormone regulation and may impair vision. Usually non-cancerous. Often treated with medication, radiation, or minimally invasive surgery.
MEDIUM RISK
🟢
NO TUMOR
Healthy brain tissue with no detectable tumor mass in the MRI scan. Normal neurological anatomy confirmed by the classifier. Routine monitoring may still be recommended by a physician.
LOW RISK
IEEE
2026

PUBLISHED & PEER-REVIEWED

NeuroScan AI is built on research published at the 2026 IEEE International Students' Conference on Electrical, Electronics, and Computer Science (SCEECS). The paper demonstrates 99.0% classification accuracy with AUC 0.999 in 10-fold cross-validation across a dataset of 16,000+ brain MRI images from three validated sources.

DOI: 10.1109/SCEECS68810.2026.11429884 → Read Full Paper
// INCEPTION V3 · PCA · NEURAL NETWORK CLASSIFIER

MRI SCAN
ANALYZER

Upload a T1 or T2-weighted brain MRI for instant AI-powered classification across four tumor categories.

MRI Image Input
Google Gemini API Key FREE
Free key at aistudio.google.com · Click Get API Key · No credit card needed · Stored in memory only
Drop MRI scan here
or click to select file
PNG · JPG · WEBP · DICOM preview · Any resolution
MRI
AXIAL
T1W
NEUROSCAN
SEQ: —
InceptionV3
PCA 95%
Neural Net
k=3,5,10 CV
4 Classes
Analysis Results
AWAITING SCAN INPUT
Upload an MRI image to begin
Preprocessing MRI image...
Extracting InceptionV3 embeddings...
Applying PCA dimensionality reduction...
Running neural network classifier...
Generating diagnostic report...
CLASSIFICATION RESULT · NEUROSCAN AI
CLINICAL RISK ASSESSMENT
Confidence Per Classification Class
No Tumor
0%
Glioma
0%
Meningioma
0%
Pituitary
0%
Clinical Observations
01
Research-Backed
Methodology from IEEE SCEECS 2026. 16K+ training images from Kaggle, BRISC, and FigShare.
02
Multi-Model Validated
Neural Net, SVM, Random Forest — all benchmarked. Neural Net achieves AUC 0.999 in 10-fold CV.
03
Not For Clinical Use
Research prototype only. Always consult a qualified radiologist or neurologist for medical decisions.
Published Research

Deep Learning Based Brain Tumor Classification Using Orange Data Analytics

IEEE SCEECS 2026
DOI: 10.1109/SCEECS68810.2026.11429884
March 2026
6 Authors · 2 Institutions

Research Team
Praveen R
Lead Author
praveenram2510@gmail.com
NIT Puducherry · Computer Science & Engineering
YOU
Monish M
Co-Author
mm0180@srmist.edu.in
SRM IST · Computational Intelligence
T R Saravanan
Co-Author
saravant1@srmist.edu.in
SRM IST · Computational Intelligence
R Archith Shri Hari
Co-Author
ar9802@srmist.edu.in
SRM IST · Computational Intelligence
Sibinath K
Co-Author
sk8381@srmist.edu.in
SRM IST · Computational Intelligence
Sumanth R
Co-Author
sr4148@srmist.edu.in
SRM IST · Computational Intelligence
Abstract

This research presents a brain tumor classification workflow using the Orange image data mining tool, covering every step from image import and visualization to machine learning evaluation. The pipeline includes data preparation, feature extraction using InceptionV3 image embeddings, and dimensionality reduction with Principal Component Analysis (PCA). Six algorithms — Random Forests, SVM, Neural Networks, Logistic Regression, Naive Bayes, and Decision Trees — are systematically trained and evaluated using k-fold (k=3, 5, 10) cross-validation. Results show the neural network model consistently achieving classification accuracy above 96% and AUC up to 0.997, reaching a peak of 99.0% accuracy and AUC 0.999 in 10-fold validation — improving on prior work by +7%.


Methodology Pipeline
01
Data Collection & Integration
Combined 16,000+ MRI images from three validated public sources. Expert radiologist annotation for every image across four clinical categories.
Kaggle: 7,023
BRISC: 6,000
FigShare: 3,064
02
Preprocessing & Augmentation
Resize to 224×224, pixel normalization. Augmentation: random rotation ±15°, horizontal flip, brightness/contrast ±20%, Gaussian noise σ=0.1.
+8% MCC
from augment.
03
Feature Extraction
InceptionV3 CNN embeddings → 2048-dimensional feature vectors per scan. Combined with handcrafted texture, contrast, variance, and mean intensity features.
2048-dim
vectors
04
PCA Dimensionality Reduction
PCA reduces 2048-dim to 100–150 components retaining 95% of original variance. Eliminates noise and redundancy. Reduces computation time by 70%.
95% variance
−70% compute
05
Model Training & Evaluation
6 classifiers trained in parallel: Neural Network (Adam optimizer), SVM, Random Forest (n=100, max_depth=10), Logistic Regression, Naive Bayes, Decision Tree.
k=3,5,10
stratified CV

Experimental Results
ModelAUCAccuracyF1 ScorePrecisionRecallMCC
Accuracy Comparison (10-Fold)
Key Findings
+7%
Accuracy Improvement
Over previous state-of-the-art approaches on equivalent brain tumor MRI classification tasks.
−70%
Compute Time
PCA reduction cuts inference computation by 70% with less than 5% variance sacrifice — essential for real-world deployment.
+8%
MCC Gain
Data augmentation alone added 8 percentage points to Matthews Correlation Coefficient — critical for minority class detection.
// THE STORY BEHIND NEUROSCAN

RESEARCH
MEETS
PRODUCT

A team of SRM IST students turned a published IEEE paper into a working AI tool that anyone can use — doctors, researchers, patients, and the curious.

OUR MISSION

Brain tumors affect millions globally. Early detection dramatically improves patient outcomes. Yet most AI research on this topic stays locked in academic papers — inaccessible to the clinicians and patients who could benefit. NeuroScan AI closes that gap.

We took our IEEE-published research on brain tumor classification and built it into a tool anyone with an MRI image can use — no medical imaging software required, no technical knowledge needed. Just upload and analyze.

We are Computational Intelligence students at SRM Institute of Science and Technology, Kattankulathur. This project started as a research paper and became something we're genuinely proud of.

Technology Stack
🧠
INCEPTION V3
Deep CNN Embedder
📊
ORANGE
Visual ML Pipeline
🔬
PCA
Dimensionality Reduction
CLAUDE AI
Vision Analysis Engine
Training Dataset
7,023
Kaggle Dataset
4 clinical classes. Widely used benchmark for brain tumor classification research.
6,000
BRISC 2025
T1 and T2-weighted MRI sequences enriching modality diversity.
3,064
FigShare
Expert-annotated tumor locations with cross-center variability.

⚠ IMPORTANT DISCLAIMER

NeuroScan AI is a research prototype built to demonstrate the findings of our published IEEE paper. It is NOT a certified medical device, NOT approved for clinical diagnosis, and must NOT be used as the sole basis for any medical decision.

This tool is intended for educational, research, and portfolio purposes. Always consult a qualified radiologist, neurologist, or specialist for brain tumor diagnosis, evaluation, and treatment planning. If you or someone you know is experiencing neurological symptoms, please seek immediate medical attention.