This project aims to detect anemia using images of the conjunctiva (the mucous membrane that covers the front of the eye and lines the inside of the eyelids). Anemia is a condition where there is a deficiency of red blood cells or hemoglobin in the blood, leading to fatigue, weakness, and other health issues. Early detection is crucial for timely treatment and management.
In this project, I utilized five different models to detect anemia from conjunctiva images. These models include:
Convolutional Neural Networks (CNN) XGBoost Classifier Logistic Regression VGG16 (Transfer Learning) ResNet50 (Transfer Learning) Models Used
-
Convolutional Neural Network (CNN) A deep learning model specifically designed for image classification tasks. The CNN model is trained to learn features from conjunctiva images that are indicative of anemia.
-
XGBoost Classifier An ensemble learning method based on gradient boosting. The XGBoost classifier is trained on features extracted from the images to predict the presence of anemia.
-
Logistic Regression A statistical model that uses a logistic function to model a binary dependent variable. In this project, logistic regression is used as a baseline model to classify images as anemic or non-anemic.
-
VGG16 (Transfer Learning) VGG16 is a convolutional neural network model pre-trained on ImageNet. Transfer learning allows us to leverage the learned features from VGG16 and fine-tune the model for the specific task of anemia detection.
-
ResNet50 (Transfer Learning) ResNet50 is another pre-trained CNN model known for its deep residual learning capabilities. Like VGG16, we fine-tune ResNet50 for our specific classification task.
The best model out of all these was the logistic regression one.
Dataset The dataset used for this project is provided by Appiahene P, Asare JW, and Donkoh E. The dataset consists of conjunctiva images from Ghana and is specifically aimed at detecting iron deficiency anemia. It is available on Mendeley Data and can be accessed through the following citation: https://data.mendeley.com/datasets/nt7r8hv2pz/