Improving Tumor Diagnosis Accuracy with CNN Based Image segmentation and Arduino Decision Support

A machine-learning algorithm that is currently having an increased growth in image segmentation is the convolutional neural network algorithm (CNN). This design is projected for classification of 3 stages of tumor types. The image is first filtered using the 2D Adaptive Median Filter (2D-AMF) to remove noise. The Contrast Limited Adaptive Histogram Equalization (CLAHE) technique is used to improve the denoised image. The Region Of Interest (ROI) is segmented from the improved image using clustering and threshold techniques. Clustering operation is done by using Fast Fuzzy C Means Clustering (FFCM) and threshold is performed using Otsu threshold (OT) Technique. The Grey Level Co-Occurrence Matrix (GLCM), a feature that calculates the occurrence of pixel pairings in certain spatial coordinates of an image, is used to precisely extract features from the ROI. Finally, obtained features are classified and manipulated by machine learning based Convolutional Neural Network (CNN) techniques. Arduino gives the classification whether the image is normal or abnormal. The decision of classifier is based on segmented output image. If the segmented output image is affected, classification using Arduino will give output as abnormal and otherwise normal.