UMamba<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Medical

U-Mamba: Hybrid CNN-Mamba architecture for medical segmentation.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → MedicalSegmentationBase<T> → UMamba<T>

For Beginners

Medical image segmentation with long-range dependencies. CT and MRI organ segmentation.

Common use cases:

  • Medical image segmentation with long-range dependencies
  • CT and MRI organ segmentation
  • 3D medical volume segmentation
  • Efficient medical AI with linear complexity

How It Works

Technical Details:

  • Hybrid CNN + Mamba (State Space Model) blocks in U-Net
  • Linear complexity for processing long-range dependencies
  • Captures both local CNN features and global SSM context
  • U-Net architecture with Mamba blocks replacing transformer layers

Reference: Ma et al., “U-Mamba: Enhancing Long-range Dependency for Biomedical Image Segmentation”, arXiv 2024.

Example

// Create a U-Mamba model for medical image segmentation with long-range dependencies
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.MultiClassClassification,
    inputHeight: 256, inputWidth: 256, inputDepth: 1, outputSize: 14);
var model = new UMamba<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new UMamba<double>(architecture,
    options: new UMambaOptions { NumClasses = 14, DropRate = 0 });
            
// Or load a pre-trained ONNX model for CT/MRI organ segmentation
var onnxModel = new UMamba<double>(architecture, "umamba.onnx");

Constructors

Constructor Summary
UMamba(NeuralNetworkArchitecture<T> architecture, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction, UMambaOptions options) Initializes UMamba in native (trainable) mode.
UMamba(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, UMambaOptions options) Initializes UMamba in ONNX (inference-only) mode.

Methods

Method Summary
MedicalSegmentationResult<T> SegmentFewShot(Tensor<T> queryImage, Tensor<T> supportImages, Tensor<T> supportMasks)
MedicalSegmentationResult<T> SegmentSlice(Tensor<T> slice) Segments a single 2D medical slice.
MedicalSegmentationResult<T> SegmentVolume(Tensor<T> volume) Segments a 3D volume slice-by-slice and aggregates the per-structure statistics.
ModelMetadata<T> GetModelMetadata() Collects metadata describing this model’s configuration.