PROJECT INSIGHTS!

Smart solutions are required in the current situation of increasing medical complexities of diseases. With each day that a new disease or virus getting discovered it is important to come up with design of systems for detection and accurate identification of pneumonia. The disease causes millions of deaths each year and the field of Medical Science requires a modern and scientifically equipped approach . The most recent kind of pneumonia is being caused due to Corona Virus .Here is an image depicting that kind.

    IMAGES OF PNEUMONIA PATIENTS FROM TEST AND TRAIN DATASET OF OUR MODEL

Technology behind the model was kept fairly simple and precise. We worked on the training of layers by adding a multitude of hidden layers and increasing parameters. Our problem statement mentions binary classification as there are only two classes of categorization – Pneumonia and Normal. 
The first step after loading data paths in the program.

This is the pre processing step of DenseNet model. The dataset size is increased to train better.

The step after pre processing is model creation.We then move on to train and test the model with sufficient number of epochs and analyse the results.

 To evaluate the best accuracy of the this model, we conducted the experiments for varying
epochs out of which 10 was found suitable for all 6 models. Hyper-parameters were tuned to increase the performance and accuracy of the model. Different accuracy was obtained by varying the algorithms, implying the various pre-processing techniques the best of them were selected. 

The graph shows CNN has best performance.

Data augmentation techniques involves refining of data and increasing the size of dataset to not only affect accuracy but also minimize the loss.

CONTRIBUTORS OF THE PROJECT:
APURVA RAJ
VISHWAS B
SANDEEP SAJAN
MRITYUNJAY SINGH

LINK TO GITHUB CODE:
https://github.com/ApurvaCodek/Pneumonia_Detection_Using_CXRS