MedCLIPOptions

Options & Configuration · AiDotNet.VisionLanguage.Encoders

Configuration options for the MedCLIP model.

Inheritance: ModelOptions → ContrastiveEncoderOptions → MedCLIPOptions

For Beginners

MedCLIP is designed for medical imaging but with a clever twist: instead of needing perfectly matched pairs of images and descriptions, it can learn from any image-text combination that describes the same medical condition. This greatly increases the amount of usable training data in the medical domain.

How It Works

MedCLIP (Wang et al., 2022) from UCSD addresses the challenge of limited medical image-text pairs by decoupling image and text inputs during contrastive learning. Instead of requiring exact image-text pairs, it uses a semantic matching loss that allows any image to be paired with any text description that shares the same medical concepts (e.g., diagnosis, anatomy).

Constructors

Constructor Summary
MedCLIPOptions() Initializes default MedCLIP options.
MedCLIPOptions(MedCLIPOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
ContrastiveLossType LossType { get; set; } Gets or sets the contrastive loss type.
DomainSpecialization Domain { get; set; } Gets or sets the domain specialization.
bool UseEntityExtraction { get; set; } Gets or sets whether to use entity extraction for concept alignment.
double EntitySimilarityThreshold { get; set; } Gets or sets the medical entity similarity threshold for soft labeling.
double SemanticMatchingWeight { get; set; } Gets or sets the weight for the semantic matching loss.
string TextModelId { get; set; } Gets or sets the reference pretrained clinical text checkpoint identifier.
string TokenizerDirectory { get; set; } Gets or sets an optional local Hugging Face tokenizer directory.
string VisionBackbone { get; set; } Gets or sets the vision backbone used by MedCLIP.
string VisionModelId { get; set; } Gets or sets the reference pretrained vision checkpoint identifier.