List of Papers Browse by Subject Areas Author List
Abstract
Ultra-low-cost smartphone-based fundus imaging offers a scalable glaucoma screening solution in resource-limited settings, yet deployment is challenged by severe acquisition noise. While short video is preferred over single-image capture in such settings, selecting diagnostically relevant frames from degraded recording remains an open problem, as existing quality assessment methods often fail under these extreme noise conditions of low-cost imaging. This work presents a domain-informed frame selection framework tailored to ultra-low-cost fundus video. The method integrates optic disc localization as a spatial quality gate with complementary spectral and structural quality metrics (retinal saturation, red-channel dominance, vessel density, and sharpness). These metrics are combined via a multiplicative fusion strategy, penalizing frames that fail in any single modality. Evaluation on a multi-center dataset of 284 smartphone-acquired videos using a MobileNetV3-based classifier shows this holistic fusion (mainly S_total) significantly stabilizes sensitivity and F1-Score compared to single-metric approaches. Fusion-based frame selection achieves an area under the receiver operating characteristic curve of 0.95 ± 0.04, closely approaching a semi-automated expert baseline (0.98 ± 0.01). These findings demonstrate that robust frame selection bridges the gap between automated and expert intervention, enabling reliable glaucoma screening on ultra-low-cost imaging platforms.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2033_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{AbaSol_Comparative_MICCAI2026,
author = { Abay, Solomon Gebru AND Asmare, Melkamu Hunegnaw AND Jacob, Julie AND Geurts, Lucca},
title = { { Comparative Analysis of Automated Frame Selection Methods for Glaucoma Detection in Smartphone-Based Low-Quality Fundus Video } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16885},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
1.This paper proposes an automated frame selection framework for ultra-low-cost fundus videos, which includes localization of the optic disc region and multiplicative fusion of four frame selection indicators.
2.This paper conducts corresponding experiments on the proposed automated frame selection framework, validating the feasibility of the method on a multi-center private dataset of smartphone-based fundus videos. Under the proposed framework, the classification performance for glaucoma achieves an AUC of 0.95, an accuracy of 0.89, a sensitivity of 0.87, a specificity of 0.90, and an F1-score of 0.86, which is very close to semi-manual methods. Statistical calculations show no significant difference between them, indicating that the proposed framework can effectively bridge the gap between automatic selection and expert intervention.
3.At the clinical application level, the automated frame selection framework proposed in this paper can assist in the deployment of ultra-low-cost smartphone-based fundus imaging. Even under challenging conditions with severe imaging noise, this framework can relatively accurately select diagnosis-relevant frames from degraded footage, addressing the shortcomings in this application domain.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The automated frame selection framework proposed in this paper integrates the optic disc localization task into the frame selection task, using the pre-trained YOLOv8n from reference [2] to detect the optic disc and form the region of interest (ROI). Meanwhile, a comparative analysis based on the optic disc detection results shows that this ROI strategy offers certain advantages.
2.The automated frame selection framework proposed in this paper performs multiplicative fusion of four frame selection indicators, accompanied by detailed visualization analysis and ablation experiments, verifying the algorithmic robustness brought by multiplicative fusion.
3.The construction and use of the dataset in this paper are relatively standardized. This private dataset was collected from two clinical centers, comprising 284 fundus videos (114 glaucoma patients, 170 healthy patients). Each video (10–17 seconds) was recorded by trained staff under mydriatic conditions. Glaucoma cases were clinically confirmed, while healthy subjects were validated through comprehensive ophthalmic examinations (visual field testing, intraocular pressure measurement, slit-lamp examination) and adjudicated by senior ophthalmologists.
4.The automated frame selection framework proposed in this paper holds value for clinical deployment and can assist ultra-low-cost smartphone‑based fundus imaging. Even under challenging conditions with severe imaging noise, this framework can relatively accurately select diagnosis‑relevant frames from degraded footage.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The algorithmic novelty of this paper is somewhat insufficient. In the proposed automated frame selection framework, the main innovations lie in the ROI strategy and the multiplicative fusion strategy. Regarding the ROI strategy, which localizes the optic disc region in fundus images, this is already a common technique in glaucoma research and is not highly innovative. Moreover, the paper employs the results of others’ work, namely the pre-trained YOLOv8n from reference [2], which further diminishes the novelty of the ROI strategy. As for the multiplicative fusion strategy—multiplying four existing frame selection indicators—this fusion method is too straightforward and simplistic, lacking algorithmic depth.
2.The text of this paper contains some minor errors. In Table 1, the metric name “Red Score” is used, which is inconsistent with the earlier “Red-Channel Intensity Score (RSC)”.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The algorithmic novelty of the automated frame selection framework proposed in this paper is limited, and the experimental completeness is slightly lacking. However, the overall experimental results of this framework are quite impressive, as its classification performance for glaucoma is very close to that of semi-manual methods. Therefore, it holds certain value for practical application and deployment.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
The core technical contribution is a multiplicative fusion of four existing image quality metrics (retinal saturation, red‑channel dominance, vessel density, Laplacian‑based sharpness), combined with an optic disc detector to gate the analysis to the region of interest. Each of these components is well established in the literature and is used here in a straightforward, plug‑and‑play manner. The fusion itself (multiplication of normalized scores) is a simple heuristic without theoretical grounding or learned adaptation. No new algorithm, model architecture, or learning paradigm is introduced.
While the application domain (ultra‑low‑cost video) is challenging, the paper does not demonstrate that its approach substantially advances the state of the art beyond careful engineering of existing tools. A comparison against even basic deep learning‑based quality assessment models (e.g., a lightweight CNN trained on the same data) is missing, making it impossible to assess whether the proposed hand‑crafted metrics offer any advantage over data‑driven alternatives.
It does not meet the bar for novelty expected at a top conference.
Review #2
- Please describe the contribution of the paper
This paper addresses an important and clinically relevant problem: automatically selecting the optimal frame from short smartphone-captured fundus videos to enable accurate glaucoma diagnosis.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The key strength of the paper lies in the methodological novelty proposed for frame selection, along with an attempt to evaluate the approach by comparing it with a semi-automated baseline.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The key weakness of this work is the limited reliability of the semi-automated baseline. In the baseline, the final frame is selected manually by the first author from the top-20 ranked candidates. This introduces two concerns: (1) the proposed method may fail to include the truly optimal frame within the top-20 candidates, and (2) the first author does not appear to be a clinical expert (e.g., an ophthalmologist), which raises questions about the validity of the final frame selection.
A more appropriate evaluation would involve an independent expert (such as a trained ophthalmologist) selecting the optimal frame, followed by a direct comparison with the frame selected by the proposed algorithm.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
Additional concerns include missing or unclear experimental details. For example, the average length of the input videos is not clearly specified (the paper only mentions “several hundred frames”), which makes it difficult to assess the practical setting.
Finally, the proposed method does not outperform the semi-automated baseline, which raises questions about its effectiveness and practical utility.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Already answered above.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
My concerns were adequately answered in the rebuttal.
Review #3
- Please describe the contribution of the paper
The primary aim of this work is glaucoma detection from ultra-low-cost smartphone fundus videos. The authors propose a domain-informed frame selection method that integrates optic disc localization with spectral and structural quality metrics. They evaluate both individual and combined quality features for frame selection, demonstrating that the fusion-based approach outperforms single-metric selection in glaucoma classification using the lightweight MobileNetV3.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
-Interesting topic: glaucoma detection using smart phone captured videos, proposing a quality based frame selection method. -Presenting several quality measures for the frame selection, then demonstrating that combining these measures gave more robust overall results.
- Manuscript is well-written and easy to follow. -Solid analysis.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
General Comments: -References numbering should be revised to be ordered based on appearance in text not alphabetically.
- Please clarify in the text whether the quality metrics were previously implemented in for retinal image quality assessment (in such case add relevant references) or are newly proposed in this paper.
- Clearly indicate whether a special camera was added to the smart phone for the video capturing. In such case, indicate its details.
- Analysis is performed on a single dataset using 5-fold cross validation. Better evaluate the method using a held-out test dataset. As well as using different datasets from different sources to assure generalizability of the presented method. (Although authors do report this as a direction for future work).
-Missing information:
- hardware specifications and tools used
- Threshold value in VE
-Minor Typos:
- Page 5: remove exclamation mark from ‘MobileNetV3-Small!’
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Glaucoma detection from smart phone videos is an interesting topic. Methods are very formulated and experiments are robust. The main limitations are that the approach was evaluated on a single dataset (with cross-validation) and it is not clearly stated whether the quality measures were previously implemented or newly introduced. Nevertheless, overall, the work represents a comprehensive and valuable foundational study.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Nothing in rebuttal led to me changing my stance
Author Feedback
We thank the reviewers (R1, R2, R3, MR1) for their constructive feedback and recognizing our experimental analysis. We address key concerns below. All missing details and typos will be corrected in the final manuscript. 1.Origin of quality metrics & methodological novelty (R1, R3, MR1) We acknowledge that the base metrics are adapted from prior literature, which will be explicitly cited. However, our core novelty is not the invention of a new optic disc (OD) detector or a single new image quality operator. Rather, the novelty lies in the task-specific, OD-gated, video-level selection formulation for ultra-low-cost fundus videos. Existing methods typically evaluate whole frames and assume stable visibility. In contrast, our framework introduces a sequential pipeline: an OD spatial gate, an OD-centered region-of-interest, per-video normalization, and multiplicative fusion. This structurally penalizes frames that fail in any single quality dimension, preventing high-frequency artifacts (e.g., glare) from spoofing the selection, which is a severe limitation empirically observed for single metrics in our analysis. The computational simplicity of this fusion is a deliberate design choice to enable edge-device deployment and low-resource scalability, effectively reducing the manual-selection gap without heavy neural architectures. 2.Reliability of the semi-manual baseline (R2, MR1) The semi-manual approach was intentionally incorporated to approximate a performance upper bound. Our key result is that the automated fusion framework eliminates the manual selection bottleneck while achieving downstream classification metrics with no statistically significant difference from this baseline. Regarding curation validity: an exhaustive ophthalmologist review of all 117,185 frames was clinically impractical. Therefore, to ensure clinical reliability, a calibration subset of 694 frames was initially evaluated. This subset was reviewed by two senior ophthalmologists to establish strict consensus criteria for diagnostic viability. Once these standards were established, the first author applied these calibrated criteria to select the final frames. Furthermore, pre-filtering the top-20 candidates using our algorithm was practically necessary to bypass the massive volume of entirely blank or severely blurred frames, making the manual review viable. 3.Dataset validation & generalizability (R3, MR1) We recognize the importance of robust evaluation. As mentioned in Section 2.1, our dataset is multi-center, comprising 284 videos collected across two distinct clinical sites. This inherently captures natural variability in data acquisition and operator handling. While our subject-level stratified 5-fold cross-validation strictly prevents data leakage and demonstrates strong internal robustness, we agree with R3 and MR1 that it does not definitively establish external generalizability. Evaluating the proposed framework on independent low-quality dataset is essential to fully solidify these findings. We will explicitly highlight external validation and the collection of a larger held-out test set as a primary direction for future work. 4.Missing or unclear implementation details & corrections (R1, R2, R3, MR1) Video length (R2, MR1): Videos range from 10 to 17 seconds captured at 30 fps (as mentioned in section 2.1), resulting in a total of 117,185 frames across the 284 preprocessed videos. We will mention this explicitly. Hardware & acquisition (R3, MR1): No special camera was used. The acquisition relied entirely on the smartphone camera combined with the GSoP adapter. We will state this in the revised manuscript. Typos & Consistency (R1, R3): We will correct the text error in Table 1.We will also remove the exclamation mark after “MobileNetV3-Small”. Reference formatting (R3): We use the official MICCAI LaTeX template, which automatically sorts references alphabetically. We will ensure strict compliance with final publisher formatting guidelines.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
This paper studies an important and clinically meaningful problem, namely automatic frame selection from ultra-low-cost smartphone fundus videos for glaucoma screening. The paper proposes a frame selection framework that combines optic disc localization with multiple spectral and structural quality cues, and the reported results suggest that the fused strategy can achieve performance close to a semi-manual selection pipeline. The work therefore has potential practical value for low-cost screening settings.
At the same time, several weaknesses remain and should be addressed clearly in the rebuttal. First, the methodological novelty appears somewhat limited, as the main contribution is largely based on combining existing quality indicators with an ROI strategy. Second, the current evaluation has an important limitation in that the semi-manual baseline depends on final frame selection by the first author from the top-ranked candidates, which makes the reliability of this reference less convincing. Third, the validation is still relatively limited, since the study is conducted on a single private dataset with cross-validation and without stronger held-out or external validation, so the generalizability of the findings is not yet fully established. In addition, several implementation and experimental details remain unclear, including aspects of the acquisition setup and some parameter settings, and the paper would also benefit from clarifying whether the quality metrics are newly introduced here or adapted from prior work.
Overall, the authors are encouraged to respond carefully to the reviewers’ comments in the rebuttal.
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
After considering the reviews and the rebuttal, I recommend acceptance. The paper should be considered mainly as an application study. Its value lies in a practical OD-gated frame selection pipeline for ultra-low-cost fundus videos, rather than a new learning model. The method is simple and interpretable, which fits the intended low-resource deployment setting. The reported results are close to the semi-manual baseline, supporting its practical usefulness.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
After considering the rebuttal and all reviewers’ comments, the overall assessment remains positive. Although some concerns remain regarding evaluation and validation, the paper addresses a practical problem with clear application value, and the results support acceptance.
