Biomedisa

Biomedisa performs semi-automatic segmentation of large volumetric images by using smart interpolation to fill sparsely pre-segmented slices for generating dense annotations used in deep neural network training.

Key Features:

• Semi-automatic segmentation: Performs segmentation of large 3D volumes using a combination of sparse user-provided labels and automated interpolation.
• Smart interpolation: Intelligently fills sparsely pre-segmented slices by considering the entirety of the underlying volumetric image data.
• Reduction of dense pre-segmentation: Decreases the need for dense manual annotations by generating intermediate segmentations from sparse inputs.
• Avoids morphological interpolation: Replaces conventional morphological interpolation approaches with data-driven interpolation across the volume.
• Improved accuracy and efficiency: Integrates comprehensive analysis of underlying image data to enhance segmentation accuracy and processing efficiency relative to traditional tools.
• Applicability to 3D imaging modalities: Applicable to a wide range of 3D imaging modalities and large volumetric biomedical datasets.
• Training-data facilitation: Produces dense annotations suitable for training deep neural networks.

Scientific Applications:

• Deep learning annotation: Generation of dense, neural-network-ready segmentations from sparse manual labels for deep neural network training.
• Large-volume segmentation: Semi-automatic segmentation of large volumetric datasets in biomedical imaging studies.
• Cross-modality volumetric analysis: Use across various 3D imaging modalities and biomedical research requiring detailed volumetric image analysis.

Methodology:

Applies a smart interpolation technique that analyzes the entire volumetric image data to fill sparsely pre-segmented slices, replacing morphological interpolation and enabling semi-automatic segmentation.