top of page
neuro image

Tumor

Glioblastoma segmentation (enhancing , necrosis)

Automatic glioblastoma segmentation using convolution neural network (영상의학과 김호성/박지은 교수님 협력연구)

그림1.png
그림2.jpg

Radiomics feature robustness analysis: transitional + deep features

Brain tumor classification (영상의학과 김호성/박지은 교수님 협력연구)

​Radiomic features and multilayer perceptron network classifier: a robust MRI classification strategy for distinguishing glioblastoma from primary central nervous system lymphoma” Sci. rep. 2019.

그림3.png

Stroke

On set time classification

DWI-FLAIR mismatch

Mismatch of visibility of an acute ischemic lesion between DWI and FLAIR

On set time classification

DWI-FLAIR mismatch

Mismatch of visibility of an acute ischemic lesion between DWI and FLAIR

그림4.png

* D. Leander Rimmele and Götz Thomalla “Wake-up stroke: clinical characteristics, imaging findings, and treatment option – an update” Frontiers in Neurology (2014)

On set time classification

Image processing

그림5.jpg

On set time classification

Infarct segmentation

  • Initial segmentation using ADC map

  1. Normalized ADC map

  2. Absolute thresholding

그림6.png

On set time classification

Ratio map generation

  • Find the mid-sagittal plane

Using the standard template with the known plane

Non-rigid registration

3D plane fitting

그림7.png

On set time classification

Feature extraction

  • Size

  • Intensity

  • Gradient

  • GLCM (gray level co-occurrence matrix)

  • GLRLM (gray level run-length matrix)

  • LBP (local binary pattern)

On set time classification

Test SVM

Using the independent test set

그림8.png

On set time classification

Classification

  • SVM (support vector machine)

With radial basis function (RBF) kernel

  • 2 binary classification tasks

Within 4.5 hours (≤ 4.5) vs. beyond 4.5 hours (> 4.5)

Within 6 hours (≤ 6) vs. beyond 6 hours (> 6)

Clot detection on GRE(신경외과 강동화 교수님 협력연구)

By analyzing the GRE sequence of the MRI with artificial intelligence, the clot, which is the cause of stroke, can be found quickly and accurately

그림14.jpg

Infarct segmentation(영상의학과 정승채 교수님 협력연구)

Results of our proposed method and various segmentation networks

Woo, Ilsang, et al. "Sementic segmentation with squeeze-and-excitation block: application to infarct segmentation in DWI." NIPS, 2017.

그림9.png

Figure. (a) b1000 image (b) ADC map (c) DenseNet (d) DenseNet+SE (e) 2D U-Net (f) 2D U-Net+SE (g) Ground truth

Table. The comparison of various segmentation models with and without SE block

그림10.png

Black blood imaging: vessel analysis

Lumen segmentation (영상의학과 정승채 교수님 협력연구)

Cascaded 3D U-net based lumen segmentation in T1 MRI images.

그림11.png

Vessel wall estimation & measurement

FWHM (full width at half maximum) based wall estimation of skeletonized lumen.

Measurements: wall thickness, wall area, wall area %, lumen diameter, lumen area.

그림12.png

Bipolar disorder: Radiogenomics Study

Bipolar disorder : Radiogenomics study (고려대학교 함병주 교수님 협력연구)

Our radio-genomics model could be used whenever a patient takes new medical imaging depending on condition and thus could be applied for the early prediction of BD. Effective prediction model using radio-genomics analysis is potentially useful for establishing biomarkers for BD and preliminary steps forward diagnosis on personalizing disorder

그림13.png
그림14.png

Brain extraction  (영상의학과 김호성/박지은 교수님 협력연구)

T1-weighted images of the different brain disease using deep convolutional neural network

image23.jpg
bottom of page