Abstract
Blood cancer's high incidence and quick progression make it a serious health issue, especially Acute Myeloid Leukemia (AML). Early diagnosis is essential to improve patient survival rates and provide efficient treatment. Traditional approaches for machine learning, like Convolutional Neural Networks (CNN), have shown effectiveness in medical image analysis, but limitations in detecting malignant features at early stages require more advanced techniques. This project proposes a hybrid model that combines the strengths of Swin Transformer and ConvNeXt architectures to enhance the accuracy and speed of AML detection in microscopic blood cell images.The model utilizes data from the AML-Cytomorphology_LMU dataset, consisting of 18,365 microscopic single-cell images labeled into 15 types of blood cells. Combining Swin Transformers' hierarchical structure with ConvNeXt's extended feature gain, the hybrid model enhances image segmentation and reliably identifies cancerous cellular features. The model reached 96% accuracy after long training and testing, significantly outperforming classic CNN techniques.
The hybrid model's great accuracy and capacity to deal with complicated image patterns highlight its potential to transform medical diagnostics. This approach might give clinicians a more precise tool for diagnosing blood cancers, such as AML, in their early stages, increasing patient outcomes and treatment efficiency.