Smart FL: meta-learning for federated blood marrow smear classification
-
By
-
N. Ilakiyaselvan
-
Srivastava Sanskar
-
D. Aarthi
-
V. Kalyanasundaram
-
July 9, 2026
-
Clinical Scorecard: Federated Learning with Meta-Learning for Enhanced Classification of Bone Marrow Smears
At a Glance
| Category | Detail |
|---|
| Condition | Bone Marrow Cell Classification |
| Key Mechanisms | Federated learning and meta-learning to enhance model adaptability and privacy. |
| Target Population | Patients with hematological disorders requiring bone marrow analysis. |
| Care Setting | Clinical laboratories and research institutions. |
Key Highlights
- Achieved 96 ± 1% accuracy in classes with low sample sizes.
- Utilized a ResNet-18 backbone for the meta-learning algorithm.
- Addressed data privacy concerns through a federated learning framework.
- Improved generalization capabilities across diverse patient populations.
- Facilitated model training without centralizing sensitive data.
Guideline-Based Recommendations
Diagnosis
- Utilize automated classification methods for accurate diagnosis of hematological conditions.
Management
- Implement federated learning to enhance model performance while preserving data privacy.
Monitoring & Follow-up
- Track cell populations over time to assess disease progression.
Risks
- Challenges in accurately classifying rare cell types due to data scarcity.
Patient & Prescribing Data
Individuals with hematological disorders such as leukemia and lymphoma.
Tailor treatments based on specific abnormalities identified in bone marrow analysis.
Clinical Best Practices
- Incorporate AI and deep learning for improved classification of bone marrow cells.
- Ensure data privacy through decentralized data management strategies.
Related Resources & Content
