Impulse Classification Network (ICN) for video Head Impulse Test

---

This research proposes the Impulse Classification Network (ICN) using 1D Convolutional Neural Network (1D CNN) that able to detect noisy data and classify human VOR impulses. ICN is a high-performance classification method that works on a patient's video Head Impulse Test (vHIT) impulse data by identifying abnormalities and artifacts. Our ICN method found actual classes of patient’s impulses with 95% accuracy. Paper link

We provide train and test python files. We created our dataset which came from the ICS goggle device.

Lateral canal test type: left side vHIT data with a) normal and b) artifact impulses

• 

• 

Four type of classes

• Normal - 1081
• Abnormal - 804
• Artifact_phase_shift - 797
• Artifact_high_gain - 1115

Total 3797 impulses

Training part

python train.py -l labels.pickle

Test part

python test.py -m ./data/yourmodel.h5 -l lables.pickle -i impulse.csv

Result of test part

We can check the VOR gain values and predict the class of each impulse.

Artifact_high_gain – 100% Normal class is detected 93.44% Artifact_phase_shift – 98.72%

Dataset _labels_ready_new.csv - classes _left_ready.csv - impulses >>> row number 1:EYE_data 2:HEAD_data, 3:EYE_data 4:HEAD_data and etc ...