Glaucoma Detection Projects 2026 — IEEE Research & Final Year Projects for BE, MTech & PhD in Bangalore
Glaucoma is the second leading cause of irreversible blindness worldwide, affecting over 80 million people. Early automated detection from fundus photographs and OCT scans is one of the most active areas in medical image analysis and deep learning research. At ProjectsatBangalore, we offer 20+ IEEE 2026 glaucoma detection projects in both MATLAB (Image Processing Toolbox, Deep Learning Toolbox, Computer Vision Toolbox) and Python (TensorFlow, PyTorch, scikit-image, OpenCV) covering the full pipeline from fundus image preprocessing to optic disc/cup segmentation, CDR computation, RNFL thickness analysis, multi-class classification, GAN-based data augmentation, Explainable AI and Federated Learning. All projects use publicly available benchmark datasets: REFUGE, RIM-ONE, ORIGA, DRISHTI-GS, HRF, DRIONS-DB, G1020 and LAG. Ideal for BE, MTech Biomedical Engineering / ECE / CSE and PhD researchers at VTU, Anna University, JNTU and NIT.
Glaucoma Project Research Areas We Cover
- CNN-based glaucoma classification — ResNet, VGG, EfficientNet, DenseNet
- U-Net / Attention U-Net optic disc and cup segmentation
- Vision Transformer (ViT / Swin-T) for fundus image classification
- Cup-to-Disc Ratio (CDR) computation and threshold-based screening
- OCT RNFL thickness map analysis for early glaucoma detection
- GAN / cGAN synthetic fundus image generation for data augmentation
- Explainable AI — Grad-CAM, SHAP, LIME for clinical decision support
- Federated Learning for privacy-preserving multi-hospital glaucoma screening
- MATLAB traditional image processing — morphological disc detection, vessel segmentation
- Multi-scale and multi-task learning for simultaneous disc, cup and vessel detection
- Transfer learning from ImageNet / EyePACS for glaucoma fine-tuning
- Hybrid CNN + SVM / Random Forest feature extraction classification
Tools, Frameworks & Libraries
Every project uses a well-documented tool stack — choose MATLAB or Python depending on your university requirement. Both paths include complete source code, trained model weights, result figures and documentation.
Python 3.x
MATLAB R2024b
PyTorch 2.x
TensorFlow / Keras
U-Net / Att-U-Net
ViT / Swin Transformer
GAN / cGAN / CycleGAN
OpenCV · scikit-image
OCT · RNFL Analysis
Grad-CAM · SHAP · LIME
Glaucoma Benchmark Datasets — 2026
All major publicly available glaucoma datasets — fundus photographs, OCT scans, disc/cup segmentation masks, CDR ground truth and glaucoma class labels — used across our IEEE 2026 projects.
Dataset 01
REFUGE (Retinal Fundus Glaucoma Challenge)
1,200 high-resolution fundus images (800 train, 400 test) with pixel-level optic disc and optic cup segmentation masks, CDR ground truth and binary glaucoma labels. MICCAI 2018 challenge gold standard — the most widely cited benchmark for glaucoma detection models.
Dataset 02
RIM-ONE (v1 / v2 / v3 / DL)
Open retinal image repository for optic nerve evaluation. RIM-ONE DL contains 313 normal and 172 glaucomatous stereo fundus images with expert-annotated optic disc and cup boundaries. Multiple expert annotations enable inter-rater reliability studies. Best for CDR and stereo depth-based projects.
Dataset 03
ORIGA (Online Retinal Fundus Image Database)
650 fundus images (482 normal, 168 glaucoma) with expert-annotated optic disc and cup boundaries from the Singapore Malay Eye Study. Widely used for CDR-based screening and optic cup/disc segmentation. Ground-truth boundaries provided by trained clinicians.
Dataset 04
DRISHTI-GS (Glaucoma Screening)
101 retinal fundus images (50 normal + 51 glaucoma) with OD/OC boundary annotations from four ophthalmologists. Soft probability maps from multiple annotators enable uncertainty-aware segmentation training. Ideal for Bayesian deep learning and attention map projects.
Dataset 05
HRF (High-Resolution Fundus)
45 high-resolution (3504 × 2336 px) fundus images with vessel, OD and background segmentation ground truth — 15 healthy, 15 diabetic retinopathy, 15 glaucoma. Used for vessel-aware optic disc detection and joint segmentation of vessels + disc. Excellent for multi-task learning projects.
Dataset 06
G1020 (Large-Scale Fundus Glaucoma)
1,020 fundus images collected from clinical settings with glaucoma diagnosis labels, optic disc bounding boxes and cup segmentation masks. Provides high class diversity and real-world noise characteristics. Suitable for large-batch deep learning classification and transfer learning projects.
Dataset 07
LAG (Large-Scale Attention Glaucoma)
5,824 fundus images (2,392 positive, 3,432 negative) with attention map labels showing regions of clinical interest marked by ophthalmologists. Specifically designed for attention-guided deep learning and weakly supervised models. Best for Class Activation Mapping (CAM) and self-attention transformer projects.
Dataset 08
DRIONS-DB (Optic Nerve Head)
110 colour fundus images with optic disc boundary delineations from two clinical experts. Captures challenging cases including peripapillary atrophy and vessels crossing the disc boundary. Used for active contour, level-set and boundary-aware CNN segmentation projects in MATLAB and Python.
20+ IEEE 2026 Glaucoma Detection Project Topics
All titles are aligned to IEEE JBHI, IEEE Access, Computers in Biology and Medicine, Medical Image Analysis and Expert Systems with Applications 2025–2026. Every project includes full MATLAB / Python source code, trained model, result figures, ROC/AUC curves, confusion matrix, IEEE base paper, university-format report, PPT and viva Q&A.
| # | IEEE 2026 Glaucoma Detection Project Title | Tool / Dataset |
|---|---|---|
| 01 | Deep CNN (ResNet-50) for Glaucoma vs Normal Classification with Transfer Learning CNN / Python | Python · PyTorch · REFUGE · ROC/AUC |
| 02 | U-Net Optic Disc and Cup Segmentation for CDR-Based Glaucoma Screening on ORIGA Dataset U-Net / Python | Python · TensorFlow · ORIGA · IoU / Dice |
| 03 | Attention U-Net for Optic Cup Segmentation — Multi-Scale Feature Fusion on DRISHTI-GS Att-U-Net | Python · PyTorch · DRISHTI-GS · Dice / HD95 |
| 04 | Vision Transformer (ViT-B/16) for Glaucoma Classification on RIM-ONE DL and REFUGE ViT / Python | Python · PyTorch · RIM-ONE · REFUGE |
| 05 | Swin Transformer for Hierarchical Retinal Feature Extraction and Glaucoma Grading Swin-T | Python · PyTorch · G1020 · REFUGE |
| 06 | GAN-Based Synthetic Fundus Image Generation to Address Class Imbalance in Glaucoma Datasets GAN / Python | Python · PyTorch · cGAN · DRISHTI / RIM-ONE |
| 07 | CycleGAN Domain Adaptation for Cross-Dataset Glaucoma Detection — REFUGE to ORIGA Transfer CycleGAN | Python · PyTorch · REFUGE · ORIGA |
| 08 | Grad-CAM Explainability for CNN Glaucoma Classifier — Clinically Interpretable Heatmaps XAI / Python | Python · Keras · Grad-CAM · REFUGE · LAG |
| 09 | SHAP-Based Feature Attribution for Glaucoma Risk Prediction from Structural OCT Parameters SHAP / Python | Python · scikit-learn · SHAP · OCT-RNFL data |
| 10 | Federated Learning for Privacy-Preserving Glaucoma Screening Across Distributed Hospital Nodes Federated | Python · PySyft / Flower · REFUGE · G1020 |
| 11 | OCT RNFL Thickness Map Segmentation and Glaucoma Classification Using 3D U-Net OCT / Python | Python · PyTorch · RNFL OCT dataset · 3D U-Net |
| 12 | Multi-Task CNN for Joint Optic Disc, Cup and Vessel Segmentation on HRF Dataset Multi-Task | Python · TensorFlow · HRF · Multi-Task Loss |
| 13 | Hybrid CNN + SVM Glaucoma Detection — Deep Feature Extraction with Classical Classification CNN+SVM | Python · Keras + scikit-learn · RIM-ONE |
| 14 | EfficientNet-B4 Fine-Tuning for Glaucoma Screening on Large-Scale LAG Attention Dataset EfficientNet | Python · TensorFlow · LAG · Weighted CE Loss |
| 15 | MATLAB Deep Learning Toolbox — AlexNet / GoogLeNet Transfer Learning for REFUGE Glaucoma Classification MATLAB · DL | MATLAB R2024b · DL Toolbox · REFUGE |
| 16 | MATLAB Image Processing — Morphological Optic Disc Detection and CDR Computation from Fundus Images MATLAB · IP | MATLAB · Image Processing Toolbox · ORIGA |
| 17 | MATLAB Fuzzy Logic + Image Processing for Glaucoma Risk Grading Based on CDR and ISNT Rule MATLAB · Fuzzy | MATLAB · Fuzzy Logic Toolbox · DRISHTI-GS |
| 18 | Active Contour (Snake) Model for Optic Disc Boundary Delineation — MATLAB Implementation on DRIONS-DB MATLAB · Contour | MATLAB · Image Processing Toolbox · DRIONS-DB |
| 19 | Multi-Label Glaucoma Severity Grading (Mild / Moderate / Severe) Using DenseNet-121 Multi-Label | Python · PyTorch · G1020 · REFUGE |
| 20 | Self-Supervised Pre-Training on Unlabelled Fundus Images Followed by Glaucoma Fine-Tuning (SimCLR) Self-Supervised | Python · PyTorch · SimCLR · REFUGE · LAG |
| 21 | Uncertainty-Aware Bayesian U-Net for Optic Cup Segmentation with Monte-Carlo Dropout Bayesian U-Net | Python · PyTorch · DRISHTI-GS · RIM-ONE |
| 22 | Weakly Supervised Glaucoma Detection Using Attention Maps and Image-Level Labels on LAG Dataset Weakly Sup. | Python · PyTorch · LAG · CAM / GradCAM |
Topics refreshed to align with IEEE JBHI, Medical Image Analysis, CBM and Expert Systems 2026 publications. Contact us for the IEEE base paper abstract and preliminary result figures for any topic above.
Glaucoma Detection Approaches — 2026
Eight core research directions, from traditional MATLAB image processing to state-of-the-art transformer and federated learning frameworks.
Approach 01
CNN Classification
Pretrained CNNs (ResNet-50, VGG-16, EfficientNet-B4, DenseNet-121) fine-tuned on fundus datasets. Binary (glaucoma vs normal) and multi-class (mild/moderate/severe) classification with class-weighted cross-entropy for imbalanced data, ROC/AUC evaluation.
Approach 02
Optic Disc/Cup Segmentation
Encoder-decoder architectures (U-Net, Attention U-Net, DoubleU-Net, TransUNet) for pixel-level optic disc and optic cup boundary delineation. Evaluated using Dice coefficient, IoU, Hausdorff Distance and CDR error on ORIGA, DRISHTI-GS and REFUGE.
Approach 03
CDR Computation & Screening
Automated Cup-to-Disc Ratio (CDR) estimation from segmented disc and cup masks. CDR threshold-based (CDR > 0.6) glaucoma screening pipeline with sensitivity / specificity trade-off analysis. MATLAB and Python implementations — from morphological thresholding to deep mask-regression networks.
Approach 04
OCT & RNFL Thickness Analysis
Optical Coherence Tomography (OCT) retinal nerve fibre layer (RNFL) thickness map segmentation and statistical deviation analysis. 3D U-Net for volumetric OCT segmentation. RNFL clock-hour and quadrant deviation maps for early structural glaucoma changes before functional loss.
Approach 05
GAN-Based Augmentation
Conditional GAN (cGAN), StyleGAN2 and CycleGAN for synthetic fundus image generation to overcome class imbalance (glaucoma-positive samples are rare). Cross-dataset domain adaptation using CycleGAN to transfer appearance between REFUGE and ORIGA without paired annotations.
Approach 06
Explainable AI (XAI)
Clinical interpretability via Gradient-weighted Class Activation Mapping (Grad-CAM), Score-CAM, SHAP (SHapley Additive exPlanations) and LIME for highlighting optic disc rim, ISNT rule regions and neuroretinal rim thinning that the model uses for its glaucoma decision — essential for FDA/CE clinical acceptance.
Approach 07
Federated Learning
Privacy-preserving glaucoma screening across multiple hospitals without sharing raw patient fundus images. FedAvg and FedProx aggregation strategies using PySyft or Flower framework. Evaluated on heterogeneous data distributions (non-IID hospital splits from REFUGE and G1020).
Approach 08
MATLAB Traditional Image Processing
Classical MATLAB pipeline: green-channel extraction, CLAHE preprocessing, morphological top-hat + bottom-hat disc localisation, active contour / level-set boundary fitting, ellipse fitting for CDR, ISNT rule verification and SVM/KNN classification from shape + texture features. Ideal for VTU project reports.
Frequently Asked Questions
Common questions about our glaucoma detection research projects in Bangalore.
What are the best glaucoma detection project topics for MTech / PhD students in 2026?
Top IEEE 2026 glaucoma detection topics include: Deep CNN (ResNet-50 / EfficientNet) Glaucoma Classification on REFUGE, Attention U-Net Optic Cup Segmentation on DRISHTI-GS, Vision Transformer for Fundus Glaucoma Grading, GAN Synthetic Fundus Augmentation for Class Imbalance, Grad-CAM Explainable Glaucoma AI for Clinical Decision Support, Federated Learning for Multi-Hospital Privacy-Preserving Screening, OCT RNFL Thickness 3D U-Net Segmentation, Self-Supervised SimCLR Pre-Training on Unlabelled Fundus Data, and MATLAB Morphological CDR Pipeline on ORIGA. Contact us for a full list with abstract and base paper options.
What datasets are best for glaucoma detection projects?
The most widely cited benchmark datasets are: REFUGE (1,200 fundus + OD/OC masks, MICCAI challenge gold standard), RIM-ONE DL (stereo fundus with multi-expert boundaries), ORIGA (650 images from Singapore Eye Study), DRISHTI-GS (101 images with 4 expert annotations and soft labels), HRF (45 high-resolution images with vessel and disc GT), G1020 (1,020 clinical images with disc/cup masks), LAG (5,824 images with ophthalmologist attention maps) and DRIONS-DB (disc boundary delineation). Dataset selection depends on your task — REFUGE for detection, DRISHTI for segmentation with uncertainty, LAG for weakly supervised and attention methods.
Should I use MATLAB or Python for my glaucoma project?
Both are fully supported. MATLAB (R2024b with Image Processing Toolbox and Deep Learning Toolbox) is preferred by students at VTU, Anna University and JNTU where MATLAB report format is expected, and for traditional image processing pipelines (CLAHE, morphological disc detection, active contour, CDR computation). Python (PyTorch or TensorFlow/Keras) is preferred for state-of-the-art deep learning projects (U-Net, Vision Transformer, GAN, Federated Learning, XAI) where you need full control over custom training loops, loss functions and pre-trained model weights from HuggingFace or timm. We deliver complete, runnable code for whichever environment your university requires.
What is Cup-to-Disc Ratio (CDR) and why is it used for glaucoma?
The Cup-to-Disc Ratio (CDR) is the ratio of the vertical diameter of the optic cup to the vertical diameter of the optic disc, measured from a fundus photograph. A normal CDR is typically ≤ 0.5; a CDR > 0.6–0.65 is clinically suspicious for glaucoma, and CDR > 0.8 is strongly indicative of advanced glaucomatous damage. Automated CDR computation from deep learning segmentation masks (U-Net optic disc and cup segmentation followed by ellipse fitting) is the most common approach in glaucoma screening algorithms. CDR is complemented by ISNT rule analysis (Inferior > Superior > Nasal > Temporal neuroretinal rim thickness) and RNFL thickness from OCT scans.
What deliverables are included in a glaucoma detection project?
Every project from ProjectsatBangalore includes: complete MATLAB (.m files) or Python (.py / .ipynb Jupyter notebook) source code with comments, dataset download and pre-processing scripts, trained model weights (.h5 / .pt / .mat), result figures (ROC curve, AUC, confusion matrix, Grad-CAM heatmaps, segmentation overlays, Dice/IoU plots), IEEE 2025–2026 base paper PDF, university-format project report (abstract, introduction, literature review, methodology, results, conclusion, references) in CBSE/VTU/Anna University format, PowerPoint presentation and viva Q&A covering glaucoma pathology, CDR, deep learning architecture choices, dataset metrics and comparison with state-of-the-art.
Can I get a glaucoma project using Explainable AI for a PhD thesis?
Yes — Explainable AI (XAI) for medical imaging is a rapidly growing PhD research area. We offer projects using Grad-CAM, Score-CAM, SHAP (SHapley Additive exPlanations) and LIME to produce clinically interpretable visual explanations of CNN and Transformer glaucoma classifiers, highlighting the optic disc rim, inferior neuroretinal rim thinning and cup-to-disc boundary regions that drive the model's prediction. We also support quantitative XAI evaluation metrics (insertion, deletion, IROF) and clinical user studies (grading agreement between ophthalmologists and AI). Full journal paper writing, statistical analysis and IEEE/Elsevier submission support is available as part of our PhD research services.
Glaucoma Project Gallery — Bangalore
Sample outputs from our glaucoma detection projects — fundus preprocessing, optic disc/cup segmentation overlays, CDR heatmaps, Grad-CAM visualisations, ROC curves and OCT RNFL thickness maps.
Optic Disc/Cup Segmentation
ROC / AUC — CNN Classifier
Grad-CAM Attention Heatmap
U-Net Cup Segmentation Mask
OCT RNFL Thickness Map
GAN Synthetic Fundus Images
Federated Learning — Hospital Nodes
MATLAB CDR Pipeline OutputStart Your Glaucoma Detection Project Today
Whether you need a straightforward MATLAB CDR pipeline for VTU, a U-Net optic disc segmentation project for MTech Biomedical Engineering, a Vision Transformer fundus classifier for an IEEE journal paper or a full Federated Learning framework for a PhD thesis — our medical imaging and deep learning experts in Bangalore will guide you from dataset preprocessing to final result analysis, report and publication support.