HierarchiRetina: A Diabetic Retinopathy Screening and Grading Cascade with Heterogeneous Mixture-of-Experts Lesion Segmentation

Md Asif Uddin; Niloy Farhan; Rafeed Rahman; Md. Golam Rabiul Alam

cs.CVOct 13, 2026

HierarchiRetina: A Diabetic Retinopathy Screening and Grading Cascade with Heterogeneous Mixture-of-Experts Lesion Segmentation

Authors: Md Asif Uddin, Niloy Farhan, Rafeed Rahman, Md. Golam Rabiul Alam

Organizations: The authors are with the Department of Computer Science and Engineering, Brac University, Dhaka, Bangladesh

Abstract

Clinical screening for diabetic retinopathy proceeds in stages, yet most automated systems grade fundus images with a single black-box network. Three gaps remain: (i) lesions differ widely in size and contrast, but segmenters typically apply the same blocks to all; (ii) ungradable images are usually excluded or treated as an extra class, and lesion-aware graders seldom test whether lesions drive the grade; (iii) staged systems often report each stage only on the images reaching it, leaving propagated errors unmeasured. HierarchiRetina, a screening, segmentation and grading cascade with inspectable stage outputs, addresses each gap. First, HSMoE-AUNet is a U-Net whose skip, bottleneck and decoder blocks are sparse mixtures of heterogeneous experts, with lesion-matched types, counts and losses; SwinHRUNetPP, a dual-encoder vessel segmenter, uses a field-of-view-restricted loss. Second, LG-DRG separates gradability from ordinal severity and lets lesion masks query the image through gated cross-attention. Third, an end-to-end evaluation counts screening misses as errors and traces each error to its stage. Pooled Dice ranged from 0.184 (microaneurysms) to 0.579 (exudates); vessel Dice was 0.731. On 58,689 held-out images from six datasets, end-to-end quadratic weighted kappa was 0.804, with referral sensitivity 0.832 at specificity 0.947, and the screening gate caused 66% of errors. In ablations, the grade ignored the masks, even with the cross-attention gate started open. On all 4,105 official DDR test images, six-class accuracy was 85.6%, above the best published six-class result we found (83.8%). Staged screening systems should be evaluated end-to-end, with stage attribution and mask ablation, before clinical deployment.

Figures & tables

Fig. 1: HierarchiRetina. Stage I stops 39,535 test images as Grade 0; Stage II segments the 19,154 images that pass, using four HSMoE-AUNet lesion models and SwinHRUNetPP; Stage III (LG-DRG) outputs Grades 1 to 4 or Grade 5 (refer). CLAHE, contrast-limited adaptive histogram equalisation; PDR, proliferative DR.

Fig. 2: Data. (a) Development and test images per grading source (log scale). (b) Grade distribution, development (dark) and test (light). (c) Lesion pools before (solid) and after (hatched) the area filter that removes near-empty masks. (d) Vessel pool and split.

Fig. 3: HSMoE-AUNet. (a) ConvNeXt-S encoder, skip processors (CBAM followed by MoE), MoE bottleneck with self-attention, MoE decoder with CBAM and deep supervision; the inset lists expert counts per lesion. (b) The router scores $N$ heterogeneous experts and mixes the best two. MLP, multilayer perceptron.

	MA	HE	EX	CWS
Expert types	4	5	7	8
Added types	–	MS	MS, TH, DoG	MS, TH, SR, TX
Experts (skips)	6/8/10	6/8/10	7/9/11	8/10/12
Experts (bottl.)	12	12	14	16
Experts (dec.)	10/8/6	10/8/6	11/9/7	12/10/8
Parameters (M)	218.3	190.5	187.8	192.7

TABLE I: HSMoE-AUNet per lesion. Experts: skips (shallow to deep) / bottleneck / decoder (deep to shallow). Added types: MS multi-scale, TH top-hat, DoG difference-of-Gaussians, SR soft region, TX texture. FTL: focal Tversky loss (false-negative weight

a

, false-positive weight

b

, exponent). Extra losses: B boundary, OD optic disc, RC region coherence. LR: learning rate.

Fig. 4: SwinHRUNetPP. (a) SwinV2-B encoder and high-resolution CNN branch, fused at $128^{2}$ and $64^{2}$ ; numbered circles connect (a) to (b). (b) Bottleneck, attention-gated decoder, refinement and topology blocks; dashed arrows are auxiliary heads. (c) All loss terms are computed inside the FOV.

Fig. 5: LG-DRG. RGB and mask streams meet in a residual cross-attention gate scaled by $\gamma$ (initialised to 0); the pooled vector feeds a gradability head and a CORN severity head with OOF-fitted thresholds. The inset summarises Section III-C . BN, batch normalisation; GAP, global average pooling.

Model	Input	AUC	Sens.	Spec.	G1 routed	FN
SwinV2-L	384	0.884	0.777	0.855	0.458	3,769
ConvNeXt V2-L (fold models)	512	0.889–0.902	0.789–0.810	0.860–0.869	0.493–0.528	3,209–3,558
ConvNeXt V2-L	512	0.904	0.809	0.868	0.528	3,214
Hybrid ensemble (7 models)	mixed	0.909	0.818	0.872	0.534	3,078
ConvNeXt V2-L (final)	768	0.929	0.854	0.887	0.645	2,459
+ hybrid, $5\times$ weight †	mixed	0.933	0.862	0.889	0.643	2,336

TABLE II: Stage I on the 58,689-image test pool at an equal routing load of 19,154 images. G1 routed: fraction of mild-DR images routed onwards; FN: false negatives. † No leakage-free set exists to set this ensemble’s threshold; reference only.

Fig. 6: Stage I on the test pool. (a) Receiver operating characteristic (ROC) curve with the validation-derived operating point

\tau

. (b) Fraction of each true grade routed onwards. (c) End-to-end QWK against the Stage I threshold; the maximum is test-selected and shown for illustration only. (d) DR cases blocked and healthy eyes admitted at each threshold. Dashed line:

\tau

Target	$n$	Thr.	Dice	IoU	Sens.	Prec.	ROC AUC	PR AUC	Dice	Sens. / Prec.	F1	sens.
			Pixel, pooled						Image	Lesion level		Image
Hard exudates	142	0.325	0.579	0.407	0.583	0.574	0.846	0.489	$0.470\pm 0.221$	0.476 / 0.473	0.475	1.000
Haemorrhages	157	0.20	0.510	0.343	0.407	0.685	0.800	0.411	$0.419\pm 0.245$	0.412 / 0.445	0.428	1.000
Cotton-wool spots	69	0.40	0.408	0.257	0.521	0.336	0.884	0.328	$0.353\pm 0.261$	0.421 / 0.306	0.355	0.971
Microaneurysms	175	0.01	0.184	0.101	0.137	0.277	0.749	0.072	$0.242\pm 0.169$	0.405 / 0.248	0.307	0.977
Vessels	30	0.46	0.731	0.576	0.915	0.609	0.966	0.852	$0.727\pm 0.043$	–	–	–

TABLE III: Stage II on held-out test splits (no lesion-free test images). Pixel metrics pooled over test pixels (inside the FOV for vessels); image Dice as mean

\pm

SD; lesion level: connected components matched at intersection over union (IoU)

\geq 0.3

(

\geq 0.2

for MA); image sens.: fraction of images whose lesion was detected. PR, precision-recall.

Fig. 7: One correctly graded test image per true grade: input with Stage I probability (top) and Stage II masks with the final output (bottom; magenta MA, red HE, yellow EX, cyan CWS, blue vessels).

	Macro-AUC	G5 AUC	QWK	Acc.	Macro-F1
Fold 0	0.942	0.999	0.771	0.758	0.767
Fold 1	0.945	0.998	0.783	0.770	0.768
Fold 2	0.938	0.997	0.754	0.745	0.751
Fold 3	0.942	0.999	0.779	0.764	0.769
Fold 4	0.939	0.999	0.769	0.756	0.763
Mean	0.941	0.999	0.771	0.759	0.764

TABLE IV: Stage III five-fold cross-validation (

n=18{,}570

). G5: Grade 5 (ungradable). Mean and SD from unrounded values.

	Original		Retrained
Mask input	QWK	$\Delta$ QWK	QWK	$\Delta$ QWK
Intact	0.7726	–	0.7444	–
All masks zeroed	0.7725	$-0.0002$	0.7446	$+0.0002$
Masks from another image	0.7724	$-0.0003$	0.7444	$0.0000$
Without MA	0.7726	$0.0000$	0.7444	$0.0000$
Without HE	0.7726	$0.0000$	0.7444	$0.0000$

TABLE V: Mask reliance of LG-DRG (fold 0,

n=3{,}714

). Original:

\gamma

starts at 0, with weight decay. Retrained:

\gamma

starts at 0.1, no weight decay, accumulation 8 (not 4). Default 0.5-threshold decode, hence the gap to Table IV (argmax);

\Delta

from unrounded QWK.

EyePACS, DDR and APTOS 2019 only (no possible Stage II overlap)
View	$n$	QWK (95% CI)	Acc.	Macro-F1
B, OOF-fitted	58,689	0.804 (0.798–0.809)	0.814	0.659
B, argmax	58,689	0.801	–	–
A, OOF-fitted	14,409	0.732 (0.722–0.743)	0.743	–
A, argmax	14,409	0.721	–	–
B, OOF-fitted	58,047	0.800	0.815	–

TABLE VI: Cascade on the test pool. View B: all images. View A: grader only, on images of true Grades 1 to 5 that reached it.

Method	Protocol	Acc. (%)	QWK
FA+KC-Net+R2 [ 7 ]	6-class, grader	83.78	n/r
Ours	6-class, end-to-end	85.63	0.913 ∗
FA+KC-Net+R2 [ 7 ]	5-class, grader	84.49	0.862
RadFuse [ 6 ]	5-class, grader	83.32	0.855
Dual-SwinOrd [ 5 ]	5-class, grader	86.54	0.904
Ours, ungradable $\to$ Grade 4	5-class, end-to-end	85.10	0.878

TABLE VII: DDR official test split, matched protocols. Published values as reported; n/r: not reported; ∗ ungradable as a sixth ordinal level.

Fig. 8: Cascade errors. (a) End-to-end confusion matrix (row-normalised, with counts); (b) attribution of the 10,907 errors to stages; (c) distance of the 3,543 severity errors from the reference.

Dataset	$n$	DR	Grade 0	Gate sens.	Gate spec.	DR blocked	QWK A	QWK B	Acc. B
EyePACS	53,576	14,043	39,533	0.837	0.883	2,288	0.721	0.785	0.812
DDR	4,105	2,225	1,880	0.931	0.969	154	0.750	0.878	0.856
APTOS 2019	366	167	199	0.994	0.945	1	0.745	0.925	0.847
FGADR	277	256	21	0.996	0/21	1	0.872	0.810	0.776
Messidor-2	262	108	154	0.889	0.818	12	0.782	0.843	0.752
IDRiD	103	69	34	0.957	0.647	3	0.747	0.830	0.660

TABLE VIII: Per-dataset results. Gate sensitivity and specificity: Grade 0 versus Grades 1 to 5; A, B: View A, View B.

Explore similar work

Aug 8, 2026

PARAGraph: Pathology-Anatomy-Aware Hierarchical Graph for Diabetic Retinopathy Grading

Diabetic retinopathy (DR) remains a leading cause of vision loss among working-age adults worldwide, making reliable severity grading clinically important. Despite strong performance, most deep models formulate DR grading as image-level classification and do not explicitly model clinically grounded evidence, such as lesion types and spatial relations. In this paper, we propose PARAGraph, a Pathology-Anatomy-Aware Hierarchical Graph framework for DR grading. PARAGraph represents each image as a three-level hierarchical graph with lesion-level nodes, intermediate category and region nodes, and global anatomical and semantic nodes. To incorporate medical priors into nodes, we construct an optic disc-fovea-anchored coordinate frame that provides a scale- and rotation-normalized retinal reference system. Within this frame, lesion nodes are encoded with category, normalized area, and anatomical coordinates. To mitigate noisy lesion segmentation, PARAGraph uses a dual-fusion strategy that introduces global visual context into a graph semantic node and a decision-level prediction branch, improving robustness when lesion evidence is unreliable. Extensive experiments on Messidor-2, APTOS, and DDR show that PARAGraph achieves consistent DR grading performance over state-of-the-art methods. Interpretability and robustness analyses further demonstrate that its predictions are clinically grounded, closely associated with lesion evidence and robust to lesion segmentation noise.

Ziyang Zhang, Yuankai Huo, Yalin Zheng +1

Vanderbilt University, USA · University of Liverpool, UK

Graph Neural Networks Diabetic Retinopathy Grading Medical Image Classification Explainable Medical Image Analysis

Ask PDF

Aug 15, 2026

Multi-Channel Feature Fusion and Monte Carlo Dropout for Uncertainty-Aware Diabetic Retinopathy Grading

Automated five-stage diabetic retinopathy (DR) grading requires more than high accuracy alone. Medical-grade deployment calls for lesion-aware preprocessing, ordinal predictions, calibrated uncertainty, and explainability to support reliable diagnostic systems. We present a unified pipeline that addresses these requirements using a Ben-Graham-green-channel CLAHE feature representation, an EfficientNetV2-L ordinal regressor, and Monte Carlo dropout for uncertainty-driven referral. Grad-CAM provides visual explanations aligned with clinically relevant lesions. The proposed method achieves a QWK of 91.31% on the APTOS-2019 official test split, placing it within the near-perfect agreement band (>80%). At a 20% referral rate, 293 of 366 images are automatically graded with a QWK of 90.40%. More complex cases are referred for specialist assessment, demonstrating a practical trade-off among grading quality, automation, and patient safety in robust, reliable, and deployment-ready medical diagnostic systems.

Saksham Kumar

Department of Computer Science & Engineering, Indian Institute of Technology Patna, Patna, Bihar, 801106, India

Diabetic Retinopathy Grading Image Classification Explainable Medical Image Analysis

Ask PDF

Mar 12, 2026

Diabetic Retinopathy Grading with CLIP-based Ranking-Aware Adaptation:A Comparative Study on Fundus Image

Diabetic retinopathy (DR) is a leading cause of preventable blindness, and automated fundus image grading can play an important role in large-scale screening. In this work, we investigate three CLIP-based approaches for five-class DR severity grading: (1) a zero-shot baseline using prompt engineering, (2) a hybrid FCN-CLIP model augmented with CBAM attention, and (3) a ranking-aware prompting model that encodes the ordinal structure of DR progression. We train and evaluate on a combined dataset of APTOS 2019 and Messidor-2 (n=5,406), addressing class imbalance through resampling and class-specific optimal thresholding. Our experiments show that the ranking-aware model achieves the highest overall accuracy (93.42%, AUROC 0.9845) and strong recall on clinically critical severe cases, while the hybrid FCN-CLIP model (92.49%, AUROC 0.99) excels at detecting proliferative DR. Both substantially outperform the zero-shot baseline (55.17%, AUROC 0.75). We analyze the complementary strengths of each approach and discuss their practical implications for screening contexts.

Sungjun Cho

Hong Kong University of Science and Technology (HKUST), Hong Kong, China

Diabetic Retinopathy Grading Vision Foundation Model Adaptation Medical VLMs Fundus Imaging Ordinal Regression

Ask PDF