Performance Analysis of the Ensemble Model in Anaemia Detection from Unmodified Smartphone-Captured Conjunctiva Images

Abstract

Anaemia is a blood-related disorder that is sometimes a harbinger of a much more severe condition like cancer. WHO recommends context-specific diagnosing methods for anaemia in the case of in-feasibility of traditional diagnosing methods, including a haemoglobin colour scale method that is based on the colour of the blood. In this manifesto, we take it one step further and diagnose anaemia using images of lower palpebral conjunctiva captured in consumer-grade cameras. Data that include lower palpebral conjunctiva images, basic details, and haemoglobin values are collected from participants. Pixel parameters of collected conjunctiva images are extracted in various colour spaces and used as input features. An ensemble learning model was developed with K-Nearest Neighbour (KNN), Random Forest and Decision Tree algorithms which showed better classification performance in classifying anaemic and non-anaemic samples. This Ensemble Model-Based Anaemia Classifier (EMBAC) performed well with both training and test sets and produced 90.91% sensitivity, 89.06% specificity, 89.69% accuracy, and area under the Receiver Operating Characteristics(ROC) curve of 0.90 for the unseen test data. EMBAC also outperformed the conventional ensemble algorithms in terms of sensitivity and accuracy. EMBAC will classify the samples more correctly irrespective of the image acquisition device under any lighting conditions without any additional hardware. This research work attempts to lay the foundation for a smartphone application to detect anaemia in real-life conditions which will enable access to a mobile healthcare instrument to diagnose anaemia for the proletariat.