Cancer Detection and Categorizing Using Convolutional Neural Network from the CT Scan Images

Abstract

With approximately 10 million deaths and 208.9 billion dollars in treatment costs in 2020, cancer has recently risen to the top of the list of killers in the globe. Additionally, more individuals are receiving cancer diagnoses each year; by 2040, there will be 27.5 million new instances of cancer annually. "Machine learning" is a subfield of artificial intelligence that employs a variety of statistical, probabilistic, and optimization techniques to enable computers to "learn" from prior knowledge and locate challenging patterns in huge, noisy, or complex data sets. Particularly well-suited for this capability are medical applications, especially those that depend on complex proteomic and genomic data. Thus, the detection and diagnosis of cancer make extensive use of machine learning. The objective of the current study is to create a deep learning (DL) instrument that can categorizes CT scan images of the kidneys and brain into a) kidney tumour, b) kidney normal, c) glioma, d) meningioma, and e) pituitary tumour (CNN). The 25000 photos in the dataset, which was gathered from Kaggle, were divided at random into training and testing data. Images were utilised for teaching 70% of the time and testing 30% of the time. To lessen skewness and enhance CNN performance, the images were reduced in size to 150*150 pixels, randomly rotated horizontally, transformed to grayscale, and had their colour ranges standardised to [0, 1]. Then, the Python modules pytorch and torchvision were used to build the CNN model. There were 3 pooling layers and 3 convolutional layers employed. For activation, the ReLU function was utilised. 50 epochs were used to test the CNN model, however PyTorch selects the number of epochs with the highest degree of testing accuracy. The actual label of the image is contrasted with the output label. Training accuracy for our model was 0.7895, and testing accuracy was 0.7561. Our study anticipates that doctors will use this tool to automate the process of finding eye diseases. This might speed up the process and encourage early detection.