Application of Machine Learning in Detecting Iron Deficiency Anemia Using Conjunctiva image Dataset from Ghana
Anemia is a global public health issue that mostly emerges as a result of a decrease in red blood cell count and is particularly prevalent in Africa. Invasive ways of detecting anemia are expensive and time-consuming. Anemia may, however, be diagnosed using non-invasive technologies such as machine learning algorithms. In our study, we compared machine learning models to detect anemia using the conjunctiva of the eyes. The main datasets consisting of the conjunctiva of the eyes were acquired using Ghana as a case study for dataset collecting. Before the study began, the ethical committees at the hospitals involved approved the collection of datasets. Also, because the participants (patients) in the study were minors, the ethical agreement was obtained from their parent(s) or guardian(s), and the purpose and objectives of the study were explained to them, along with the advantages of the health services. Before the participants were enrolled in the data collection, their parent(s) or guardian(s) gave their consent. Furthermore, the ethics and consent committee of the University of Energy and Natural Resources, Ghana approved the start of this experiment. Furthermore, patients' or participants' names and faces were not shown or exposed during image capture, rendering their identification unknown
Steps to reproduce
All photographs were taken by the laboratory personnel, and the Hb Values of patients, age, sex, and a remark (anemic or non-anemic) based on the Hb Value taken were uploaded by technicians or medical laboratory officers using standard high-quality cameras with a minimum of 12MP. To have accurate exposure to the images of the conjunctiva of the eyes, the lower eyelid was gently pushed back with the thumb and perhaps index finger. Furthermore, to minimize exaggerated shine effects caused by the picture quality, which greatly impacts detection or classification by the models, the cameras' spotlights were turned off when the photos were photographed. The datasets were collected from the undermentioned hospitals located in Ghana; Komfo Anokye Teaching Hospital at Kumasi, Bolgatanga Regional Hospital at Bolgatanga, Kintampo Municipal Hospital at Kintampo, Ahmadiyya Muslim Hospital at Techiman, Sunyani Municipal Hospital at Sunyani, Manhyia District Hospital at Kumasi, Ejusu Government Hospital at Ejusu, SDA Hospital at Sunyani, Nkawie-Toase Government Hospital at Nkawie-Toase and Holy Family Hospital at Berekum. The dataset for this study focuses on young children aged five and under since the World Health Organization studies show that forty-two percent of young children under the age of five and forty percent of pregnant and expectant women are anemic. The threshold triangle approach was used to extract the ROI of the pictures of the conjunctiva of the eyes; that is, ROI differentiation for binary after the background was generated was inferred in the dataset collecting procedure. As the Region of Interest is produced from the counter of the photos, which is the conjunctiva of the eyes, the "triangle algorithm for thresholding" was applied. Following that, the images were classified as anemic or non-anemic based on the corresponding comments provided by the laboratory Hb result. In the procedure of augmenting the images to expand the size of the original datasets from 710 images (426 are anemic images while 284 are non-anemic images) to 4260 datasets after the image augmentation of because of the augmentation, all photos were labelled "anemic" and "non-anemic," with a number matching to each. The reason that all participants are under the age of six, images of the conjunctiva of the eyes may be tiny, as compared to that of adults’ images of the conjunctiva of the eyes.