Preliminary COVID19 Diagnosis Based on Cough Recording Classification

Preliminary COVID-19 Diagnosis Based on Cough Recording Classification Group number: <54> Author: <Jiayu Zhao> <Zhuoran Liu>

COVID 19 Outbreak People all around the world are suffering from this pandemic.

Challenges we are facing ● ● No vaccine available Tests needed to be done on site Pressure on the medical system Miss best treatment time for infected people Confirmed cases keep increasing daily…. .

How does machine learning help in this case? 1. A laboratory free detection method. 2. Cheap. 3. easy.

Literature search ● ● Cough detector ○ acts as a filter before the diagnosis engine and is capable to distinguish sound COVID-19 diagnosis

Literature search

Details on the dataset Negative cough: 1287 Positive cough: 89 Optional: ESC-50

Details on feature extraction used Mel Frequency Cepstral Coefficients (MFCC) 1. Python library: Librosa 2. Figure of cepstrum 3. A set of eigenvectors

Details on the model used CONV 2 d: 2*Conv 2 d layers (64, (2, 2), padding='same', activation = "relu") 2*Conv 2 d layers (32, (2, 2), padding='same', activation = "relu") 2*Maxpooling layers ((3, 3), strides=(2, 2), padding='same') Flatten layers 3*dense layers(256>64 ->2) (Dropout layers are testing)

CONV 1 D 4*Conv 1 d layers( kernel_size=4, strides=1, padding="same) filter: 20 ->18 ->15 ->10 Global. Max. Pooling 1 D Batch. Normalization Dense(2, activation='softmax')

Observations MFCC observations Figure 1: Waveform of the signal Figure 2: MFCC graph of the signal MFCC demonstrates clearer distinction with different color denotation.

Result Figure 1: cough dataset shape (X_train, X_val, X_test) Figure 2: CNN model summary

Further items to be completed ❖ Dataset resample ❖ Conv 1 d testing ❖ Dropout layer implementations

Reference [1] COVID-19 Dashboard by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University (JHU). Accessed on: April 23, 2020. [Online]. Available: https: //coronavirus. jhu. edu/map. html [2] Imran A, Posokhova I, Qureshi H N, et al. AI 4 COVID-19: AI Enabled Preliminary Diagnosis for COVID-19 from Cough Samples via an App[J]. ar. Xiv preprint ar. Xiv: 2004. 01275, 2020.

DEMO TIME