Issuu

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 09 Issue: 05 | May 2022

p-ISSN: 2395-0072

www.irjet.net

Detection of Covid-19, Skin Cancer and Malaria using AI and Image Processing Snehima Gupta1, Vivek Shukla2, Mandeep Singh Narula3 Student, Dept. of ECE, Jaypee Institute of Information Technology, Noida, India Student, Dept. of ECE, Jaypee Institute of Information Technology, Noida, India 3 Assistant Professor, Dept. of ECE, Jaypee Institute of Information Technology, Noida, India ---------------------------------------------------------------------***--------------------------------------------------------------------2. CONVOLUTION NEURAL NETWORKS Abstract - Owing to the burgeoning population and with it, 1

the upswing in the number of diseases, the synergy of technology and the healthcare sector has become a pressing need. It's essential to integrate technology like АI to trace, track and detect critical diseases and make frameworks for upcoming ones. There have been several developments in machine learning technology especially, to use it to aid the doctors. It decreases the time taken to diagnose and speeds up the treatment phase.

Convolution neural network is a network of various hidden layers between input and output which help in feature extraction and classification based on probabilistic models. Any basic CNN model contains the following layers:

A large number of data sets consisting of imageries are formed for various ailments and are made open for comprehensive research. These data sets form the basis of AI solutions and deep learning algorithms. The solution we propose uses АI frameworks and data sets available for SARS- CoV 2 also known as Covid19, Malaria and Skin Cancer to classify and label them as healthy or affected. Key Words: Machine Learning in disease detection, Image Classification, CNN, AI, VGG 19, ResNet 18, Inception V3, DenseNet 201

1. INTRODUCTION The proposed solution is a streamlined and efficient way of screening three critical diseases using image processing. The solution leverages fine-tuned version of four advanced CNN models namely, ResNet 18, VGG -19, Inception V3 and DenseNet 201 to identify and label the provided chest X-Ray scans/ blood smear/ skin lesions from healthy ones. This opportunity has also been wielded to compare these four models on their accuracy, speed and complexity. [1]

Impact Factor value: 7.529

Input layer, its where the input image matrix is given.

•

Convolution layer, it basically breaks down the large image matrix into smaller segments and features.

•

ReLU or activation Layer, the rectifier linear unit assigns zero to all negative values in the matrix so that the predictions are better.

•

Pooling layer, it helps to reduce the dimension of these matrices, there are three types of layers min max and average.

•

Then there’s the fully connected layer which classifies the objects into the required categories along with their closeness probability.

•

SoftMax or logistics layer depending upon the classification type.

And finally, the output layer which shows the most probable match. [3]

Research has been done on Convolution Neural Networks, Image Processing, ResNet 18, VGG -19, Inception V3 and DenseNet 201 to provide a viable solution that can help in improving the medical technician’s ability to detect a particular disease and play a vital role in the reduction of fatalities due to the same. [2] The proposed approach has also taken the intra class resemblances and inter class variations in the color, texture, location, scale and look of the sample image into account for better identification and results. The above has been completed in three phases, the first one used ResNet 18, the second one uses VGG-19 and the final one uses Inception V3 and Densenet 201.

•

Fig 1. Basic CNN Architecture

ISO 9001:2008 Certified Journal

Page 2210