About this trial
To establish a multimodal fundus image report generation model to realize an interpretable system for multiple fundus diseases, multimodal image analysis, diagnosis, and treatment decision automatic reporting based on weakly labeled training data. Construct an interpretable feature fusion network for the clinical and imaging features of fundus lesions, and we hope to extract new imaging markers that can predict the occurrence and progression of various fundus lesions at an early stage, and ultimately verify them in real clinical data, further providing possible directions for exploring the molecular mechanisms of refractory fundus lesions, and may also provide new ideas for the precise prevention and treatment of fundus lesions.
Eligibility criteria
This trial accepts healthy volunteersQualifiers
Disease group: All multimodal fundus examination images containing fundus lesions, examined from January 2011 to December 2023, including fundus photography, as well as OCT, OCTA, FFA, ICGA, B-ultrasound and corresponding imaging reports. Images could be either clear or unclear, reports are either complete or incomplete.
Normal group: All multimodal fundus examination images without fundus lesions, examined from January 2011 to December 2023, including fundus photography, as well as OCT, OCTA, FFA, ICGA, B-ultrasound and corresponding imaging reports. Images could be either clear or unclear, reports are either complete or incomplete.
Disqualifiers
Disease group: 1. The image has serious quality problems; 2. The diagnostic report lacks key information.
Normal group: 1. The image has serious quality problems; 2. The diagnostic report lacks key information.
Trial population
The clinically collected fundus examination images and corresponding reports, including various categories of fundus lesions or normal findings, obtained from examinations conducted at the Zhongshan Ophthalmic Center.
Trial design
Cohort
Retrospective
Treatments tested in this trial
With fundus diseases
Other interventionWith fundus diseases
Without fundus diseases
Other interventionWithout fundus diseases
Treatment groups
Trial outcomes
Primary outcomes
Fundus Fluorescein Angiography (FFA) images with corresponding report
Fundus Fluorescein Angiography(FFA) images with corresponding report were collected. FFA allows dynamic observation of changes in retinal blood vessels and lesions.
Indocyanine Green Angiography (ICGA) images with report
Indocyanine Green Angiography (ICGA) images with corresponding report were collected. ICGA allows dynamic observation of changes in choroidal blood vessels and lesions.
Fundus photography with corresponding report
Fundus photography images with corresponding report were collected. Fundus photography provides observation of morphological manifestations of the retina, the retinal blood vessels, the optic nerve, as well as lesions on the retina.
Optical coherence tomography (OCT) images with corresponding report
Optical coherence tomography (OCT) images with corresponding report were collected. OCT images provides observation of changes in retinal thickness, morphology and manifestations of lesions in each retinal or choroidal layer, as well as lesions in the macular area and optic nerve
Secondary outcomes
Bilingual Evaluation Understudy (BLEU) analysis
Bilingual Evaluation Understudy (BLEU) measures how closely a machine-generated text matches the reference texts to quantify the similarity between the generated text and reference texts.
Metric for Evaluation of Translation with Explicit Ordering (METEOR) analysis
Metric for Evaluation of Translation with Explicit Ordering (METEOR) considers precision, recall, alignment, and includes stemming and synonymy to quantify the similarity between the generated text and reference texts.
Recall-Oriented Understudy for Gisting Evaluation (ROUGE) analysis
Recall-Oriented Understudy for Gisting Evaluation (ROUGE) focuses on recall and measures the overlap of n-grams, contiguous sequences of n items (words, characters, or symbols) extracted from a given sample of text, between the generated and reference texts to quantify the similarity between the generated text and reference texts.
Pearson correlation analysis
To determine the degree of linear relationship between human evaluations and automated assessments.
Sponsors and contacts
Click on the lead sponsor to view all of their trials.
This trial is not recruiting at the moment. You can still explore other options: