Adoption of Shallow Neural Networks in Pneumonia Classification

A CNN contains a several layers each receiving input from the preceding layer. The final layer of the CNN flattens the image into a column and then determines which features most correlate to a particular class. In Kenya, pneumonia accounts for 16% of the total number of deaths in children under the age of five and ranked as the second leading cause of death with this age group. Currently radiologist examine Chest X-rays images under luminous light to check for pneumonia. This study proposes the use of a five-layer shallow neural network for image classification. The model records an accuracy of 91% and AUC scores of 90% with the type I error and type II errors are 14% and 5% respectively on secondary data. On local tests data, the model recorded an ROC of 90% and Type 1 error rate and type II error rate was 11.7 % and 7.4% respectively