Abstract:
Effective leukemia detection in resource-limited environments mostly in developing country like Rwanda necessitates the use of advanced machine-learning algorithms. This study, titled "Design and develop Leukemia Detection Using Deep Learning" addresses the critical need for timely and accurate leukemia diagnosis that is crucial for successful treatment.
Acknowledging the challenges posed by traditional diagnostic methods, such as reliance on expert pathologists and time-consuming procedures as it should take more than two weeks by culturing the samples, the study has focused on leveraging advanced machine-learning techniques. Despite the absence of a local dataset in Rwanda, this research has utilized open datasets that have been prepared with images obtained from Taleqani Hospital(bone marrow laboratory) in Tehran, Iran. This dataset has 4961 training images divided into two classes: 2483 images from healthy patients and 2478 images from patients affected by blood cancer. We tested the model with a total of 1240 images, with 620 from each class, all with a resolution of 320x240 pixels. The study's findings underscore the effectiveness of the deep learning system compared to conventional diagnostic techniques. Performance evaluation metrics demonstrate its ability to enhance diagnostic accuracy and efficiency, thus presenting a significant advancement in leukemia detection. Moreover, the study suggests the integration of this system across district hospitals in Rwanda to address the limited access to leukemia diagnosis, especially considering the sole hospital capable of performing such tests. By doing so, the study aims to increase diagnostic capacities and alleviate strain on healthcare resources. The completed study offers valuable insights and practical solutions to improve leukemia diagnosis not only in Rwanda but also in similar resource-limited environments globally.
