Robust Classification of Acute Leukemia Subtypes Using CNN-Based Feature Fusion and Whole-Cell Image Analysis | ||
| IJCI. International Journal of Computers and Information | ||
| Articles in Press, Accepted Manuscript, Available Online from 18 November 2025 | ||
| Document Type: Original Article | ||
| DOI: 10.21608/ijci.2025.410828.1206 | ||
| Authors | ||
| Hadeel Farag Saber* 1; Noura A. Semary2; Khalid Amin3; Sondos Magdy4 | ||
| 1Department of Information Technology, Faculty of Computers and Information, Kafrelsheikh University, Kafrelsheikh, 33516, Egypt | ||
| 2Inf. tech. dept. , Information and computers faculty, Menofia university | ||
| 3Information Technology dept., Faculty of Computers and Information, Menoufia University, Egypt | ||
| 4Information technology dept., Faculty of computers and information, Menofia university. | ||
| Abstract | ||
| Acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML) are serious hematologic cancers. They impact the bone marrow, blood, and lymphatic system. While ALL and AML cells have some similarities, their treatments differ. Timely and accurate diagnosis is vital due to the rapid progression of these diseases. Specialists often encounter it challenging to distinguish between myeloid and lymphoid cells. This can lead to time-consuming tasks and risks of misdiagnosis, which can be life-threatening. To address this, we present a computer-aided diagnosis (CAD) method for identifying myeloid and lymphoid cells accurately and efficiently. Given the similarities between AML and ALL cells, we extracted whole cell lines, including nuclei and cytoplasm, from the background. Myeloid and lymphoid cells can be distinguished using nuclear morphology and the ratio of nuclear to cytoplasmic volume. We focused on color and morphological features from cell images, along with deep features from an efficient CNN, enhancing classification accuracy. After fusing features, we used a CNN model for classification. The model achieved outstanding results: 99.91% accuracy, 99.83% sensitivity, 99.91% F1 score, and 100% for precision, specificity, and area under the ROC curve. These results show the model’s strong performance in classifying leukemia, outperforming both traditional and recent techniques. This suggests it has enormous potential for wider use in medical diagnostics. This study utilized two publicly available datasets: the ALL-IDB2 dataset for ALL and the TCIA AML dataset for AML. Due to their different sources, images were initialized during preprocessing. | ||
| Keywords | ||
| Acute Lymphoblastic Leukemia; Acute Myeloid Leukemia; Deep Learning; CNN; Medical Imaging | ||
|
Statistics Article View: 1 |
||
