MAIA (Medical AI Aggregator) COVID-19 Detection

The first case appeared in Wuhan in December 2019. The new coronavirus belongs to the Coronaviridae family and is a single-stranded RNA positive-strand virus. The virus is highly infectious, infects via droplets and contact, and mainly infects the lower respiratory tract. It has an incubation period of 1-14 days. The symptoms are not necessarily obvious. Patients may experience fever, dry cough, tiredness, mouth and nose sensory disorders.  It is known that viral spikes mainly contact ACE2 on mucosal cells for cell membrane fusion, thereby entering a large number of viral RNA and protein shells in cells, assembling them by high-profile bodies and sending them out of the cell for the next wave of infection. COVID-19 is currently pandemic and is spreading throughout the world. Therefore, the detection of COVID-19 is an urgent issue.

Dataset information
Source Label
Stanford ML Group CheXpert Dataset
Joseph Paul Cohen, COVID-19 image data collection
COVID-19 (+)
1. Stanford ML Group CheXpert Dataset
2. Chest X-Rays tuberculosis from India
COVID-19 (-)

Data Information

Image type: Chest X-ray
Image size: 298 x 294 – 4248 x 3480

(Up Left) Healthy chest X-ray from Stanford dataset, (Up Right) COVID-19 patient chest X-ray from Github: ieee8023/covid-chest x-ray-dataset
(Down Left) Disease chest X-ray from Stanford dataset, (Down right) Tuberculosis chest X-ray from Tuberculosis dataset

Data Training and Testing Information
Dataset / Quantity Training Set
  Testing Set   
CheXpert Healthy 4455/4000 1109/1000
CheXpert Disease
COVID-19 (+) 50/35 22/14
  • The model validation set is automatically and randomly split by MAIA in the training stage.
  • COVID-19 Training and Testing data are split by patient ID.

Hardware Information
CPU: i7-8750H  

GPU: RTX-8000*2 96GB
Software: MAIA

Medical AI Aggregator (MAIA) Auto Deep Learning Technology
Deep learning is one of the fields of machine learning. Muen team has created a software-Medical AI Aggregator(MAIA) that has the ability to learn the relation between data and labels, and produce prediction models based on it's learning. Apart from traditional machine learning skills, Muen's technology achieves higher accuracy and versatility. Hundreds and thousands of images can be automatically processed in a short period of time to obtain the prediction model.
                                                               Fig2. Medical AI Aggregator Interface


Stage/ Metrics AUC Precision Recall Specificity F1 Score Accuracy
Testing 0.9500 0.8249 0.8606 0.8926 0.8416 0.8672

In model Evaluation, MAIA achieved 86.7% accuracy, 82.4%, precision, 89.3%recall, 89.3% specificity, 84.2% f1_score.
**Healthy cases from CheXpert Dataset, Disease cases from CheXpert Dataset and Tuberculosis dataset, COVID-19 cases from github: ieee8023/covid-chestxray-datase

MAIA Software classified Chest X-ray between healthy, different diseases and COVID-19 data by utilizing only two RTX 8000 GPU. MAIA not only runs with low memory consumption hardware but also output a robust model by a small amount of COVID-19 dataset.

To apply MAIA in clinic detection, it is essential to rely on professional medical practitioners to operate MAIA and thereby perform diagnosis and treatment. We look forward to fighting off COVID-19 together.