Klasifikasi Eritrosit Pada Thalasemia Minor Menggunakan Fitur Konvolusi dan Multi-Layer Perceptron

Abstract

Thalassemia blood disorder is a condition that can affect the blood's ability to function normally and can lead to erythropoiesis. In this blood disorder, there are nine types of abnormal erythrocytes, namely elliptocytes, pencils, teardrops, acanthocytes, stomatocytes, targets, spherocytes, hypochromic and normal. At present, thalassemia examination is carried out using Hb electrophoresis and is done manually so it will be subjective and take a long time. Therefore, this research implements the Convolutional Neural Network (CNN) and Multi-Layer Perceptron (MLP) algorithms. This study aims to determine the performance of convolution features as image feature extraction and MLP as an image classification method and then implemented on NVIDIA Jetson Nano. The convolution features used in this study apply the CNN VGG16 architecture. Then model learning is carried out on 7245 data by configuring hyperparameters. The best accuracy with the hyperparameter configuration is a batch that is 16, the epoch is 400, the learning rate is 0.0001, the dropout1 layer is 0.1 and the dropout2 layer is 0.1. From this configuration it produces optimal accuracy at 96.175%. In the following, the model that has been made is then implemented on the NVIDIA Jetson Nano as a mobile media to be applied to the medical world resulting in an average prediction speed for each class of 48.330 seconds. The obtained performance time and accuracy are suitable for use by medical personnel to predict the class of abnormal erythrocytes.