Title: MICCAI2018

Authors: MICCAI

Segmentation

Highlights

(1) BESNet: Boundary-Enhanced Segmentation of Cells in Histopathological Images Hirohisa Oda

for the detection and semantic segmentation of cells on histopathological images

#####two strategies for enhancing the boundaries of cells:

(1) skip connections of feature maps: the feature maps from the boundary decoding path are concatenated with the decoding path for entire cell segmentation.

(2) adaptive weighting of loss functions: an adaptive weighting of the loss for entire cell segmentation is performed when boundaries are not enhanced strongly, because detecting such parts is difficult.

#####Network structure of BESNet: BESNet has two decoding parts, Boundary Decoding Path (BDP) and Main Decoding Path (MDP). Feature maps in BDP are concatenated with MDP. Loss function for MDP is weighted by BDP output.

#####loss: Boundary-Enhanced Cross-Entropy (BECE) Loss

#####Results: Segmentation accuracy was evaluated by Dice index, precision, and recall.

The mean Dice index representing segmentation accuracy was 74.0%.

Probability threshold t was set to 0.05, 0.10, ··· , 0.95 for FROC evaluation.

The Dice index, precision, and recall of the proposed method were 71.4±31.9, was 0.50. The scores of the true positives (TPs) were computed for the highest 81.2 ± 32.9, and 67.2 ± 31 (when threshold t was 0.5)

####SPNet: Shape Prediction Using a Fully Convolutional Neural Network S. M. Masudur Rahman Al Arif

The proposed deep fully convolutional neural net- work learns to predict shapes instead of learning pixel-wise classification

#####Dataset dataset consists of 124 training images and 172 test images containing 586 and 797 cer- vical vertebrae, respectively

The manual annotation for each of the training vertebrae is converted into a signed distance function (SDF).

To convert the vertebral shapes into an SDF (Φ), the pixels lying on the manually anno- tated vertebral boundary curve have been assigned zero values.

#####Methodology

#####loss:

#####Results: Hausdorff distance

####Deep nested level sets: Fully automated segmentation of cardiac MR images in patients with pulmonary hypertension Jinming Duan

a novel and accurate optimisation method for segmentation of cardiac MR (CMR) images in patients with pulmonary hypertension (PH).