dosma.models.StanfordQDessUNet2D

class dosma.models.StanfordQDessUNet2D(input_shape, weights_path, force_weights=False)[source]

2D U-Net model trained on the 2021 Stanford qDESS Knee dataset.

This model segments patellar cartilage (“pc”), femoral cartilage (“fc”), tibial cartilage (“tc”), and the meniscus (“men”) from quantitative double echo steady state (qDESS) knee scans. The segmentation is computed on the root-sum-of-squares (RSS) of the two echoes.

There are a few weights files that are associated with this model. We provide a short description of each below:

  • qDESS_2021_v1-rms-unet2d-pc_fc_tc_men_weights.h5: This is the baseline

    model trained on the 2021 Stanford qDESS knee dataset (v1.0.0).

  • qDESS_2021_v0_0_1-rms-pc_fc_tc_men_weights.h5: This model is trained on the RSS

    2021 Stanford qDESS knee dataset (v0.0.1).

  • qDESS_2021_v0_0_1-traintest-rms-pc_fc_tc_men_weights.h5: This model

    is trained on both the train and test set of the 2021 Stanford qDESS knee dataset (v0.0.1).

Examples

# Create model based on the volume’s shape (SI, AP, 1). >>> model = StanfordQDessUNet2D((256, 256, 1), “/path/to/weights”)

# Generate mask from root-sum-of-squares (rss) volume. >>> model.generate_mask(rss)

# Generate mask from dual-echo volume de_vol - shape: (SI, AP, LR, 2) >>> model.generate_mask(de_vol)

__init__(input_shape, weights_path, force_weights=False)

Methods

__init__(input_shape, weights_path[, ...])

build_model(input_shape[, weights_path])

Builds a segmentation model architecture and loads weights.

generate_mask(volume)

Segment tissues.

Attributes

ALIASES

sigmoid_threshold