Detecting red blood cell morphology changes in iron deficiency by deep learning artificial intelligence
Iron Deficiency has a high prevalence globally including Malaysia. In uncomplicated patients, serum ferritin and red cell parameters are sufficient for diagnosis. However, ferritin is an acute-phase protein that becomes elevated in response to inflammation, complicating the diagnosis in the...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
IIUM Press
2025
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/123364/2/123364_Detecting%20red%20blood%20cell.pdf http://irep.iium.edu.my/123364/ https://journals.iium.edu.my/revival/index.php/revival/article/view/452 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Iron Deficiency has a high prevalence globally including Malaysia. In uncomplicated patients, serum ferritin and red cell parameters are sufficient for diagnosis. However, ferritin is an acute-phase protein that becomes elevated in response to inflammation, complicating the diagnosis in the presence of coexisting chronic diseases. Many studies on image processing provide the opportunity to develop an artificial intelligence model for the diagnosis of iron deficiency anaemia (IDA). However, most of them onlyfocus on a single smear image which is not sufficient to make a conclusive clinical report on the diagnosis of IDA. Therefore, a new approach using whole smear slide imagesto perform pathological classification is more appropriate to assist in the diagnosis of IDA in the presence of coexisting thalassaemia and chronic kidney disease. This study aims to develop a deep learning model to detect red blood cell (RBC) morphological changes due to iron deficiency. Three study groups (Iron deficiency anaemia, Anaemia due to Thalassemia trait and Anaemia of chronic disease (CKD) ) are defined by Malaysia Clinical Practice Guidelines. The data was collected from Sultan Ahmad Shah Medical Centre records from 2017 to 2022. Images of peripheral blood smears were analysed using a slide scanner. Pathological red blood cells were manually selected as samples. 80% of the sample is used to train a deep learning (DL) model and the remaining 20%is used to test the DL model. model. The deep learning model was able to detect RBC morphology changes due to IDA with 92% sensitivity and 94% specificity in the presence of coexisting pathologies. |
|---|