MétaCan
Menu
← Back to cohort
Record W7009098341

Détection d'anomalies dans les photographies du segment antérieur de l'oeil par apprentissage profond non supervisé

2024· other· fr· W7009098341 on OpenAlexaboutno aff

Bibliographic record

VenuePolyPublie (École Polytechnique de Montréal) · 2024
Typeother
Languagefr
Field
Topic
Canadian institutionsnot available
Fundersnot available
KeywordsEthnographyFace (sociological concept)Social analysis
DOInot available

Abstract

fetched live from OpenAlex

RÉSUMÉ: Le diagnostic des pathologies oculaires menant à la perte de vision constitue un enjeu économique et social majeur au Canada. En raison des contraintes liées au manque de personnel spécialisé, notamment dans les régions éloignées, et du manque de ressources financières, la téléophtalmologie offre une solution prometteuse en combinant imagerie médicale et analyse à distance. Largement fondée sur l’imagerie du fond d’oeil, qui bénéficie d’un riche corpus d’études scientifiques, elle nécessite cependant des équipements spécialisés et coûteux, freinant ainsi son déploiement à grande échelle. Face à ces limitations, l’imagerie du segment antérieur de l’oeil se présente comme une alternative intéressante. Plus accessible et économique, elle permet l’identification de pathologies grâce à des photographies prises à l’aide d’une lampe à fente. Ce mémoire se concentre sur le développement d’une méthode de détection d’anomalies non supervisée pour les photographies de la conjonctive. En s’appuyant sur des approches par reconstruction, l’objectif est d’identifier toute structure anatomique sortant de la définition de la normalité apprise. À cet effet, nous avons d’abord constitué une base de données adaptée, puis développé SiamAAE, un modèle d’apprentissage profond innovant qui combine reconstruction et auto-distillation surmontant les biais liés à l’apprentissage de l’identité observés dans les modèles de la littérature, tout en conservant les performances en détection d’anomalies de ces dernières. Cette étude marque une première étape dans l’exploration des méthodes non supervisées de détection d’anomalies appliquée aux photographies du segment antérieur de l’oeil. Elle vise à promouvoir la téléophtalmologie et à faciliter la détection précoce des pathologies oculaires, contribuant ainsi à une meilleure prise en charge médicale. ABSTRACT: Vision loss is a major economic and social challenge in Canada, often caused by undiagnosed or untreated ocular pathologies. imited access to specialized personnel, particularly in remote areas, and financial constraints have made teleophthalmology a promising solution. This approach combines medical imaging with remote analysis, relying primarily on retinal imaging, a well-researched field in the scientific community. However, the high cost and specialized nature of the required equipment remain significant barriers to widespread adoption. Given these limitations, anterior eye segment photography offers an attractive alternative. It is more accessible and cost-effective, enabling the detection of pathologies only using slit lamps. This project focuses on developing an unsupervised anomaly detection method in conjunctival images. By leveraging reconstruction-based approaches, the aim is to identify anatomical structures that deviate from the learned definition of normality. To support this work, a tailored database was created, and SiamAAE, an innovative deep learning network, was developed. SiamAAE combines reconstruction and self-distillation approaches, addressing identity-related biases found in existing models while maintaining high performance in anomaly detection. This study serves as an initial step in developing unsupervised methods for detecting anomalies in anterior eye segment photographies. It aims to advance teleophthalmology and support the early detection of ocular pathologies, ultimately improving medical care in Canada.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.003
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: none
GenreCandidate signal: Methods · Consensus signal: none
Teacher disagreement score0.005
Threshold uncertainty score0.017

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0010.003
Meta-epidemiology (narrow)0.0010.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0020.001
Science and technology studies0.0000.000
Scholarly communication0.0010.001
Open science0.0010.001
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0050.002

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.009
GPT teacher head0.220
Teacher spread0.211 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designSimulation or modeling
Domainnot available
GenreMethods

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2024
Admission routes1
Has abstractyes

Explore more

Same venuePolyPublie (École Polytechnique de Montréal)→French-language works237,207→