Related Experiment Video
Updated: Jun 12, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.7K
BSNEU-net: Block Feature Map Distortion and Switchable Normalization-Based Enhanced Union-net for Acute Leukemia
Rabul Saikia1, Roopam Deka2, Anupam Sarma3
1Department of Electronics and Communication Engineering, National Institute of Technology Meghalaya, Shillong, India. p21ec003@nitm.ac.in.
Journal of Imaging Informatics in Medicine
|September 25, 2024
Summary
This study introduces BSNEU-net, a deep learning framework for acute leukemia detection. The model achieved over 99% accuracy on novel and combined datasets, aiding rapid illness identification.
Area of Science:
- Hematology
- Medical Imaging
- Artificial Intelligence
Background:
- Acute leukemia, including acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML), involves rapid proliferation of immature white blood cells.
- Deep learning (DL) and artificial intelligence (AI) offer potential for enhanced diagnostic accuracy and efficiency in medical sciences.
- Accurate and timely diagnosis of acute leukemia is crucial for effective patient treatment and outcomes.
Purpose of the Study:
- To propose a novel deep learning framework, BSNEU-net, for the accurate detection of acute leukemia from blood smear images.
- To address challenges in DL model generalization and overfitting by incorporating specific architectural components.
- To evaluate the performance of the proposed framework on both a newly created dataset and a combined heterogeneous dataset.
Main Methods:
- Development of the BSNEU-net framework, featuring 4 Union Blocks (UB) with union convolution for feature extraction.
- Integration of block feature map distortion (BFMD) to minimize overfitting and enhance generalization.
- Inclusion of switchable normalization (SN) layers to improve model convergence and overcome batch normalization limitations.
Main Results:
- The BSNEU-net model achieved high accuracy, reaching 99.37% on a novel dataset of 2400 images (ALL, AML, healthy).
- On a heterogeneous dataset of 2700 images (combined public datasets), the model attained 99.44% accuracy.
- Comparative analysis demonstrated the superiority of the proposed BSNEU-net methodology over existing schemes.
Conclusions:
- The proposed BSNEU-net framework demonstrates exceptional performance in detecting acute leukemia.
- The novel architectural components, BFMD and SN, contribute to improved model generalization and accuracy.
- This DL-based approach shows significant promise for assisting clinicians in the rapid and accurate diagnosis of acute leukemia.
