Covid19-RXDX
​
Complementary Covid19 Diagnosis Solution ideal for Medical Doctors in Hospital Urgencies or Primary Care that allows an immediate detection on high risk of Covid19 patients based on Medical Images analisys with Artificial Intelligence techniques such as Machine Learning and Deep Learning.
The Model has been training with images from different databases.
It's impact will be spectacular and it will help to stop Covid19 spread in Hospital and Residences and will protect hundreds of thousands healthcare workers and the millions of most vulnerable citizens.
Who we are?
01
Engineers
Aeronautical, computer, electronic, industrial and telecommunications engineers
02
Medical Doctor
Medical Doctor expert in innovation,in Big Data and AI health projects
03
Investigadores
Researchers in Artificial Intelligence
Our History
OUR STORY
Our Base Code comes from research and development within the academic environment of the doctoral program started at UCM and continued in 2020 in the UGR's 'FisyMat' doctoral program.
An ad hoc algorithm has been developed with Convolutionary Artificial Neural Networks, which mixes modern techniques of "Transfer Learning" with others of Data Augmentation, Regularizers, etc. to train X-ray images of the Chest (also CT) of patients and diagnose Covid19.
A team was formed for the "#VencerAlVirus" Hackathon organized by the Community of Madrid (Spain) and for future Hackathons.
​
HWe have been working with the different public datasets available
​
kaggle.com/paultimothymooney/chest-xray-pneumonia/data
OUR VISION
Currently our model validation tests with 500 Images has achieved an accuracy of 98% and high specificities and sensitivities above 95%, by applying fine-tuning methods to the techniques used in Transfer Learning
However, before entering production, it is necessary to extend the training with a greater number of images and carry out the corresponding validations in Hospitals.
We believe that it is essential for Europe to lead an initiative to create a processing infrastructure as well as a repository of chest X-ray and CT images of patients diagnosed with Covid19, which will allow groups of international researchers to advance in the knowledge of the disease, its diagnosis and treatment.
This initiative will allow us to face new healthcare challenges caused by viruses in the future.
TECHNOLOGY
We use 'Open Source' tools with Python, Tensorflow and Keras to create the models using Convolutional Neural Networks and Transfer Learning.
Results
Our artificial neural network model based on supervised Machine and Deep Learning methods resulting of applying Transfer Learning techniques in addition to specifics techniques of regularizers and data augmentation applying on fine tuning optimizers over a 1232 Images samples distributed on 50% labeled by Covid-19 and the other Non Covid-19 associated with normal and other types pneumonias, training afeter 30 epochs result in a final performance of 100% Accuracy, 100% Precision, 100% specificity and 100% Sensitivity plus other relevant metrics shown on "figure8_valora_Params_metricas_binarias" here adjoint. We think we are on the right way to improve and generalise better our model avoiding any potential biases included in our dataset images.
