Mini-COVIDNet

Mini-COVIDNet : Efficient Light Weight Deep Neural Network for Ultrasound based Point-of-Care Detection of COVID-19

Please cite this work if you use any codes in your work:

Navchetan Awasthi, Aveen Dayal, Linga R. Cenkeramaddi, and Phaneendra K. Yalavarthy, "Mini-COVIDNet : Efficient Light Weight Deep Neural Network for Ultrasound based Point-of-Care Detection of COVID-19," IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control (Special issue on Ultrasound in COVID-19 and Lung Diagnostics) 2021 (in press). [doi: 10.1109/TUFFC.2021.3068190]

Mobile network based models are proposed for making smaller models for COVID-19 detection and compared with state of the art techniques for ultrasound imaging. We compared our models with other state of the art techniques such as POCOVID-Net and compared our models in terms of size, number of parameters as well as the various figures of merit. All the models were run for focal loss as well for the experiments.

Contributors :

Navchetan Awasthi, Aveen Dayal, Linga R. Cenkeramaddi, and Phaneendra K. Yalavarthy

Datasets :

The dataset utilized in this work is available at: https://github.com/jannisborn/covid19_pocus_ultrasound/tree/master/data Please use the same instructions to get the data for research purposes and do cite the relevant work.

Models used for the comparisons

This repository contains the notebooks for all the codes which were run for all the models proposed and compared.

The various models compared here can be given as:

• COVID-CAPS :

This architecture has been initially used for identification of COVID-19 from chest X-ray images. It consists if convoltutional layers, and capsule layers in the architecture.

• POCOVID-Net :

Here, a VGG-16 network pretrained architecture was used for the detection of COVID-19 for ultrasound images.

• Mini-COVIDNet :

Here, a modified mobilenet architecture was used and shown to perform better for ultrasound images. It consists of depthwise convolution and separable convolution instead of normal convolution and shown to perform better.

• Mini-COVIDNet (focal loss) :

Here, a modified mobilenet architecture was used and shown to perform better for ultrasound images. It consists of depthwise convolution and separable convolution instead of normal convolution and shown to perform better using the loss function involving the focal loss.

• MOBILENetV2 :

Here, a modified mobilenetv2 architecture was used for the detection of COVID-19 in ultrasound images.

• NASNETMOBILE :

Here, a modified NasNetMobile architecture was used for the detection of COVID-19 in ultrasound images.

• ResNet50 :

Here, a modified ResNet50 architecture was used for the detection of COVID-19 in ultrasound images.