Classification and Prediction of Human Blood Cells Using Artificial Intelligence and Advanced Image Processing Techniques

Background: Accurate blood cell classification is essential for diagnosing and monitoring blood disorders. Manual blood evaluation is cumbersome and subject to disagreement among specialists, which can negatively impact diagnostic reliability. Objectives: This study aims to develop an automated deep learning framework for accurate classification of major blood cell types, especially basophils, red blood cells, and bone marrow cells, to enhance the accuracy and efficiency of clinical diagnosis. Patients and Methods: A set of publicly available, high-resolution blood smear images obtained from a specific patient cohort with distinct genetic properties was analyzed, with standardized preprocessing applied to address variance. Multiple AI-based classification strategies were developed, and all models were evaluated on an independent test set using overall accuracy, precision, recall, and F1 score. Results: Wavelet scattering combined with an SVM delivered the strongest overall performance, surpassing both the custom CNN and ResNet variants. It achieved a near-perfect separation of basophils and erythroblasts and only occasional confusion with myeloblasts. These results highlight the sensitivity of the wavelet scattering method to subtle morphological differences in blood cells. Conclusion: This study highlights how machine learning-based image analysis techniques can reliably and accurately classify blood cells, reducing the need for the subjective manual interpretation that characterizes traditional microscopy. There is potential for increasing the accuracy of early diagnosis and simplifying patient treatment plans for hematological disorders by integrating these automated systems into standard clinical practice. Keywords: Hematological Diagnostics, Blood Cell Classification, Wavelet Scattering Transform, Transfer Learning.