Automatic Brain Midline Detection based on Segmentation Network

Congress:

ECR 2020

Poster Number:

C-03025

Type:

Scientific Exhibit

Keywords:

Neuro, CNS, Computer applications, Emergency, CT, Computer Applications-Detection, diagnosis, Haemorrhage, Retrospective, Diagnostic or prognostic study, Performed at one institution

Authors:

J. Wang¹, S. Wang², K. Liang², X. Li²; ¹Tianjin/CN, ²Beijing/CN

DOI:

10.26044/ecr2020/C-03025

DOI-Link:

https://dx.doi.org/10.26044/ecr2020/C-03025

Fig. 1: Qualitative comparison between baseline(w/o alignment), baseline, proposed...

Fig. 2: midline shift in different types of hemorrhage

Fig. 3: Comparing to the gold standard, our estimated maximum distance error and area...

Purpose

To present an automatic brain midline detection using alignmnet, image segmentation and dynamic programming techniques.

Methods and materials

The proposed models are evaluated on an in-house dataset and a subset of the public dataset CQ500 with 877 and 235 non-contrast head CT scans respectively. We provides a novel framework of automated midline detection for computer aided decision support. The proposed framework firstly estimates the transformation matrix to align an input CT image into the standard space. Then, the aligned image is processed by a midline network integrated with the CoordConv Layer and Cascade AtrousConv Module to obtain the probability map. The Coord-Conv is...

Results

The proposed models are evaluated on an in-house dataset and a public dataset CQ500. For the in-house dataset, we collected 877 non-contrast head CT scans with 5-mm slice thickness from three hospitals. The dataset is randomly split into a train/val/test set of 708/87/82 stacks and the number of scans with severe midline shift is 207/44/42 respectively. For the public dataset CQ500, we exclude the health CT scans and choose 235 CT scans (53 severe midline shift) with around 5-mm slice thickness. One senior radiologist marks...

Conclusion

We present a segmentation based brain midline detection network to quantify midline shift which could evaluate the severity of patients and indication of operation precisely for neurosurgical emergency treatment.

Personal information and conflict of interest

J. Wang; Tianjin/CN - nothing to disclose K. Liang; Beijing/CN - nothing to disclose S. Wang; Beijing/CN - nothing to disclose X. Li; Beijing/CN - nothing to disclose

References

1.Liu R, Li S, Su B, Tan CL, Leong TY, Pang BC, et al. Automatic detection and quantification of brain midline shift using anatomical marker model.Comput Med Imaging Graph. 2014;38:1-14 2.Chang PD, Kuoy E, Grinband J, Weinberg BD, Thompson M, Homo R, et al. Hybrid 3d/2d convolutional neural network for hemorrhage evaluation on head ct.American Journal of Neuroradiology. 2018;39:1609-1616 3.Lee H, Yune S, Mansouri M, Kim M, Tajmir SH, Guerrier CE, et al. An explainable deep-learning algorithm for the detection of acute intracranial haemorrhage from...