MétaCan
Menu
← Back to cohort
Record W6990407495

Développement d'outils d'intelligence artificielle pour la caractérisation de lames histologiques marquées par nanoparticules plasmoniques

2024· other· fr· W6990407495 on OpenAlexaboutno aff

Bibliographic record

VenuePolyPublie (École Polytechnique de Montréal) · 2024
Typeother
Languagefr
Field
Topic
Canadian institutionsnot available
Fundersnot available
KeywordsAutomatic test equipmentMachine epsilonFeature matching
DOInot available

Abstract

fetched live from OpenAlex

RÉSUMÉ: Introduction : En 2023, près de 29 400 Canadiennes furent diagnostiquées avec un cancer du sein. L’ensemble des patientes a subit une ou plusieurs biopsies servant à faire un test de détection de protéines appelé immunohistochimie (IHC). Le test permet d’établir le plan de traitement optimal pour chaque patiente. Malheureusement, cette méthode est subjective et dépend du pathologiste qui évalue l’échantillon. Le test induit donc un manque de concordance interobservateur important lors de l’évaluation de l’expression d’une protéine d’intérêt. Le groupe de Pr. Meunier a développé une solution de quantification de protéines qui passe par l’utilisation de nanoparticules plasmoniques (NPs). Ces particules qui ciblent les protéines d’intérêt apparaissent ensuite comme des billes brillantes qu’il est possible de compter. Cela donne une mesure quantitative, mais implique qu’il faut compter les NPs attachées à l’échantillon. Cependant, la difficulté de détection des NPs par des algorithmes naïfs limite le potentiel de la méthode. Objectif : Développer des approches de détection de NPs par intelligence artificielle (IA). La réalisation de cet objectif permettra de valider l’hypothèse selon laquelle l’apprentissage machine est plus propice à la détection de NPs qu’un algorithme non entraîné. Méthodologie : L’approche choisie fut de d’abord développer une banque de données de lames marquées par NPs en respectant les conditions d’acquisition d’images en pathologie. Ensuite afin de générer une annotation des images, la stratégie employée fut d’identifier le centre des NPs en utilisant une méthode de microscopie à plus haute résolution. Ainsi, différents algorithmes de segmentation sémantique de type UNet ont pu être entraînés à détecter les NPs sur les images de la banque de données. Résultats : Cette approche a permis de déterminer les paramètres d’entraînement adéquat pour ce type de tâche de détection, puis d’effectuer une détection automatique avec une meilleure performance (FScore supérieur) que celle des algorithmes non entraînés. La comparaison des performances des différents modèles a permis une réflexion sur la pertinence des mécanismes d’attention dans le contexte de cette application ainsi que sur l’utilisation de ces algorithmes pour identifier des objets de très petite taille dans des images. Cette amélioration de la détection de NPs permettra d’augmenter la plage dynamique de cette méthode de quantification de protéines afin de mieux déterminer les plans de traitements de patients atteints du cancer du sein. ABSTRACT: Introduction: In 2023, nearly 29 400 Canadian women were diagnosed with breast cancer. Every single one of these patients had a biopsy of their tumor taken to undergo a protein detection test called Immunohistochemistry (IHC). The results of this test were then used to establish the optimal treatment plan for the patient. Sadly, the current method is subjective and extremely reliant on the pathologist’s evaluation of the sample. This yields error rates in the evaluation of patients that can go up to 30% for samples in which a protein is weakly expressed. Pr. Meunier’s group at Polytechnique Montréal developped a protein quantification solution that uses Plasmonic Nanoparticles (NPs). These particles are used as probes that target the desired proteins. The optical properties of these NPs allow them to appear as bright colorful beads that can then be counted. This enables a relative quantification of protein expression which comes with the task of counting a tremendous amount of NPs on each sample. To achieve this, naïve algorithms have been used, although, their performances are limited due to the inherent variability of the task. The poor performance of these algorithms limits the application potential of this method. Objective: Develop nanoparticle detection solutions using AI. The accomplishment of the above objective would allow us to validate the hypothesis that AI is more suitable than naïve algorithms for NP detection. Methodology: The suggested method was to first develop a database of images of Nanoparticle-stained samples. Then, the center of each NP in the database was annotated using a higher resolution microscopy method. Once constructed and annotated, this database was used to train different semantic segmentation algorithms using UNet style architectures. These algorithms could then be used to detect NPs in different samples. Results: The suggested approach allowed us to establish the optimal training parameters for this type of detection task. The resulting algorithms were able to detect nanoparticles in new samples with a higher overall performance than naïve algorithms. The comparison between different models yielded by various training parameters allowed us to discuss the relevance of attention mechanisms in this context. The improvements brought by these algorithms will increase the dynamic range of this protein quantification method, which would allow us to better establish breast cancer patients’ treatment plans.

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)0.003
metaresearch head score (Gemma)0.007
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Bench or experimental · Consensus signal: Bench or experimental
GenreCandidate signal: Empirical · Consensus signal: none
Teacher disagreement score0.003
Threshold uncertainty score0.018

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0030.007
Meta-epidemiology (narrow)0.0010.001
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0010.001
Science and technology studies0.0000.001
Scholarly communication0.0030.002
Open science0.0020.001
Research integrity0.0020.001
Insufficient payload (model declined to judge)0.0030.002

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 head0.018
GPT teacher head0.260
Teacher spread0.242 · 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 designBench or experimental
Domainnot available
GenreEmpirical

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→