Pneumonia and Covid Detection using Ct scan through Auto Encoders

Abstract

With the seasonal effects on the human body, doctors are not able to differentiate between the
pneumonia and covid because most of the symptoms are similar in both the cases from
second wave of covid. Existing approaches utilized chest X-rays for classification but those
are not accurate because in X-rays cannot expose all the details clearly. So, the proposed
model uses CT scan even though it is expensive for accurate classification and getting even
low resolution details in clear format. The virus enters the upper respiratory system in
pneumonia but either due to the less immunity power or negligence over the routine cold and
cough, the infection spreads over the lungs. This infection has impact for conversion into
COVID. The proposed model identifies level of infection by processing the CT scan images
using the auto encoders because auto encoding mechanism is efficient for extracting the both
high level and low level details. The very thin line difference between the covid and
pneumonia is observed through the bottle neck layers of the encoder. Instead of traditional
CNN’s, utilization of proposed model reduces the memory utilization because of the
representation in compressed format. It also increases the accuracy +2.5% i.e, 97.5 but it is
95% in traditional approaches.