Application of Convolutional Neural Network Image Analysis and Machine Learning to Basic Blood Tests for Intelligent
Yuki Horiuchi1,2, Mendamar Ravzanaadii1,2, Jing Bai2
1Department of Clinical Laboratory Medicine, Juntendo University Graduate School of Medicine, Tokyo, Japan.
International Journal of Laboratory Hematology
|September 11, 2025
Summary
A new deep learning system for blood cell image recognition and diagnosis shows high accuracy. Combining image analysis with complete blood count data, this diagnostic assist tool shows promise for routine clinical use.
Area of Science:
- Hematology
- Artificial Intelligence
- Medical Diagnostics
Background:
- Developed an automated deep learning system for peripheral blood cell image recognition.
- Constructed a diagnostic assist system integrating image recognition data with complete blood count (CBC) data.
Purpose of the Study:
- Evaluate the clinical performance of the image recognition deep learning system (DLS) and the diagnostic assist DLS in routine examinations.
Main Methods:
- Trained the image recognition DLS on over 1.4 million peripheral blood cell images to classify 14 cell types and 24 morphological characteristics.
- Combined image recognition DLS data with CBC data from an automated hematology analyzer to create the diagnostic assist DLS.
- Evaluated performance on over 128,000 images from healthy subjects and various hematological malignancy cases.
Main Results:
- The image recognition DLS achieved high accuracy (97.3%-99.9%) for 14 blood cell types and >90% for 11 morphological characteristics.
- Accurate detection of blast cells and classification of malignant lymphocytes were observed.
- The diagnostic assist DLS demonstrated high performance in differentiating Myelodysplastic Syndromes (MDS) with an AUC of 0.99.
Conclusions:
- The diagnostic assist DLS, integrating morphological image recognition with CBC parameters, shows significant potential as a clinical diagnostic tool.
