cs.CVMar 6, 2026

OPTED: Open Preprocessed Trachoma Eye Dataset Using Zero-Shot SAM 3 Segmentation

Authors: Kibrom Gebremedhin, Hadush Hailu, Bruk Gebregziabher

Organizations: Department of Computer Science, Mekelle University, Mekelle, Ethiopia · Department of Computer Science, Maharishi International University, Fairfield, IA, USA · Signal Technologies, Germany

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Trachoma remains the leading infectious cause of blindness worldwide, with Sub-Saharan Africa bearing over 85% of the global burden and Ethiopia alone accounting for more than half of all cases. Yet publicly available preprocessed datasets for automated trachoma classification are scarce, and none originate from the most affected region. Raw clinical photographs of eyelids contain significant background noise that hinders direct use in machine learning pipelines. We present OPTED, an open-source preprocessed trachoma eye dataset constructed using the Segment Anything Model 3 (SAM 3) for automated region-of-interest extraction. We describe a reproducible four-step pipeline: (1) text-prompt-based zero-shot segmentation of the tarsal conjunctiva using SAM 3, (2) background removal and bounding-box cropping with alignment, (3) quality filtering based on confidence scores, and (4) Lanczos resizing to 224x224 pixels. A separate prompt-selection stage identifies the optimal text prompt, and manual quality assurance verifies outputs. Through comparison of five candidate prompts on all 2,832 known-label images, we identify "inner surface of eyelid with red tissue" as optimal, achieving a mean confidence of 0.872 (std 0.070) and 99.5% detection rate (the remaining 13 images are recovered via fallback prompts). The pipeline produces outputs in two formats: cropped and aligned images preserving the original aspect ratio, and standardized 224x224 images ready for pre-trained architectures. The OPTED dataset, preprocessing code, and all experimental artifacts are released as open source to facilitate reproducible trachoma classification research.

Figures & tables

Fig. 2: Overview of the OPTED preprocessing pipeline. Raw eyelid photographs are processed through SAM 3 text-prompt segmentation, background removal, bounding-box cropping with 5% padding, horizontal alignment, and Lanczos resizing to 224×224224\times 224 pixels.
Fig. 3: Representative samples from the final OPTED dataset ( 224×224224\times 224 pixels). Top row: Normal (healthy conjunctiva). Middle row: TF (trachomatous inflammation, follicular). Bottom row: TI (trachomatous inflammation, intense).
Fig. 4: Step-by-step visualization of the OPTED pipeline on three sample images (two Normal, one Trachoma). From left to right: raw photograph, SAM 3 mask overlay, background removed, bounding-box crop with horizontal alignment, and final 224×224224\times 224 output.
PromptDet.MissScoreStdArea
red tissue inside eye99.8%60.8530.06928.2%
inner surface of eyelid>> 99.9%10.8460.08024.0%
red lining inside eyelid98.7%360.7370.07626.7%
membrane under eyelid99.5%140.8050.08226.3%
inner surface of eyelid with red tissue99.5%130.8730.06929.8%
TABLE I: Prompt comparison across all 2,832 known-label images.
Fig. 5: Radar chart comparing the five candidate prompts across five metrics: detection rate, mean confidence, consistency (1 – std), mask area ratio, and fewest misses. The winning prompt P5 (“inner surface of eyelid with red tissue”) dominates on confidence, consistency, and mask coverage.
Fig. 6: Visual comparison of SAM 3 masks from the five candidate prompts on three sample images (two Normal, one Trachoma). Blue overlay indicates the predicted mask; yellow contours delineate boundaries. The selected prompt P5 (green border) provides the most complete coverage of the tarsal conjunctiva.
Fig. 7: PSNR and SSIM comparison of four interpolation methods for resizing cropped eyelid images to 224×224224\times 224 . Lanczos achieves the highest PSNR (39.16 dB) and SSIM (0.9713), indicating best preservation of fine tissue detail.
 NormalTFTI
Processed images2,48732421
Mean confidence0.8700.8880.902
Final image size224×224×3224\times 224\times 3
FormatPNG
TABLE II: OPTED dataset statistics.
Fig. 8: Class distribution of the OPTED dataset, showing the breakdown into Normal, TF (trachomatous inflammation, follicular), and TI (trachomatous inflammation, intense) categories.
Fig. 9: Distribution of SAM 3 confidence scores across the full dataset, separated by class. Dashed lines indicate per-class means. Trachoma images achieve slightly higher confidence due to the visually distinct inflamed tissue.
 TrainValTestTotal
Normal1,7413733732,487
TF2274948324
TI153321
Total1,9834254242,832
TABLE III: Train/validation/test split by class.
More in this topic

Ophthalmic Image Analysis

See all 42 papers

Explore similar work

CardsList
Mar 18, 2026

Eye image segmentation using visual and concept prompts with Segment Anything Model 3 (SAM3)

Previous work has reported that vision foundation models show promising zero-shot performance in eye image segmentation. Here we examine whether the latest iteration of the Segment Anything Model, SAM3, offers better eye image segmentation performance than SAM2, and explore the performance of its new concept (text) prompting mode. Eye image segmentation performance was evaluated using diverse datasets encompassing both high-resolution high-quality videos from a lab environment and the TEyeD dataset consisting of challenging eye videos acquired in the wild. Results show that in most cases SAM3 with either visual or concept prompts did not perform better than SAM2, for both lab and in-the-wild datasets. Since SAM2 not only performed better but was also faster, we conclude that SAM2 remains the best option for eye image segmentation. We provide our adaptation of SAM3's codebase that allows processing videos of arbitrary duration.

Diederick C. Niehorster, Marcus Nyström

Lund University Humanities Lab & Dept. of Psychology, Lund University, Lund, Sweden · Lund University Humanities Lab, Lund, Sweden

Medical Image SegmentationPromptable Image SegmentationVision Foundation ModelsText-Guided Image SegmentationVideo Object Segmentation
AskPDF
Sep 20, 2026

Comparative Performance and Parameter-Efficient Adaptation of DINOv2 for Active Trachoma Classification

Automated grading of conjunctival photographs could reduce the cost and variability of trachoma prevalence surveys, but the relative value of modern pretrained visual representations, lightweight feature adaptation, and training-objective design has not been established under a common protocol. This study presents a controlled evaluation for binary classification of Trachomatous Inflammation-Follicular (TF) versus Normal using 1,546 images from the public UCSF/Lietman collection. Images are processed using the OPTED pipeline for zero-shot tarsal-conjunctiva segmentation, alignment, cropping, and standardization. We first compare six pretrained backbones using a common classification pipeline and then evaluate four lightweight adaptation mechanisms on DINOv2 ViT-B/14. Under stratified five-fold cross-validation, DINOv2 with Efficient Channel Attention (ECA) and focal-plus-center loss achieved 91.66 +/- 0.97% accuracy, 90.69 +/- 1.10% macro-F1, and 96.06 +/- 0.71% AUC. ECA introduces only five learnable parameters while matching the performance of substantially larger alternatives. Objective ablation further showed that ECA did not consistently improve plain DINOv2 across loss functions; the lowest-variance 91.66% accuracy was obtained with cross-entropy plus center loss. Overall, the fine-tuned DINOv2 representation provided most of the predictive performance, while ECA offered a highly parameter-efficient refinement whose effect depended on the training objective. The resulting workflow provides a reproducible benchmark for active trachoma image classification.

Kibrom Gebremedhin, Hadush Hailu, Bruk Gebregziabher +1

Department of Computer Science Mekelle University Mekelle, Ethiopia · Department of Computer Science Maharishi International University Fairfield, IA, USA · Signal Technologies Germany +1

Channel AttentionVision Foundation Model AdaptationMedical Image ClassificationParameter-Efficient Fine-Tuning
AskPDF
May 4, 2026

Scientific Domain Knowledge Improves Vision-Language Fundus Models

Vision-language models hold considerable promise for ophthalmology, but it remains unclear which training data source best conveys expert domain knowledge. Existing ophthalmic models are trained on fixed text templates, medical reports, or general biomedical literature, sources that have never been compared under matched conditions. To include domain-specific literature in this comparison, we present PubMed-Ophtha, a hierarchical dataset with high domain density of 102,023 panels with their subcaptions from 15,842 open-access articles in PubMed Central. We then finetuned identical CLIP models on each source, using a general biomedical literature model as baseline, and found that domain-specific literature achieved the best average performance across 110 clinical tasks, reaching a mean linear probing AUROC of 88.63% ahead of medical reports (85.68%). Restricting the dataset to fundus images, to the image count of the medical report dataset, or to articles unrelated to the evaluation datasets did not reduce performance, indicating that the gains likely stem from domain density. We release the dataset, the finetuned models, and the full generation pipeline.

Verena Jasmin Hallitschke, Carsten Eickhoff, Philipp Berens

1Hertie Institute for AI in Brain Health, University of Tübingen, Tübingen, Germany · Institute for Applied Medical Informatics, University of Tübingen, Tübingen, Germany · 2Tübingen AI Center, Tübingen, Germany

Medical VLMsHealthcareMedical DiagnosisVision-Language PretrainingFundus ImagingMedical Image Benchmarks+1
AskPDF