AI-Driven Bone and Marrow Segmentation on FLT-PET/CT: Technical Multi-organ Validation in AML and HCT

Area of Science:

• Nuclear Medicine
• Oncology
• Medical Imaging

Background:

• [18F] 3'-deoxy-3'-fluorothymidine positron emission tomography (FLT-PET) assesses cellular proliferation for acute myeloid leukemia (AML) and stem cell engraftment monitoring.
• Accurate quantification of bone marrow proliferation using FLT-PET/CT is challenging due to manual segmentation limitations and lack of clinical validation for automated tools.
• Existing automated tools primarily focus on solid tumors, necessitating specific validation for skeletal FLT-PET/CT in hematologic malignancies.

Purpose of the Study:

• To evaluate a deep learning-based whole-body segmentation and quantification platform for FLT-PET/CT in AML patients undergoing hematopoietic stem cell transplant (HCT).
• To assess the accuracy and efficiency of an AI tool in segmenting cortical and trabecular bone marrow compartments and organs.
• To validate the AI tool's performance against manual segmentation for reliable marrow quantification.

Main Methods:

• Deep learning whole-body segmentation applied to FLT-PET/CT scans from 20 refractory AML patients post-transplant.
• Quantification of five representative regions of interest (spleen, liver, T6, L1, L3) and separate cortical-trabecular marrow compartments.
• Comparison of automated AI measurements with manual segmentation for correlation and agreement analysis.

Main Results:

• Strong agreement (r > 0.98) between automated and manual measurements for count data, confirming AI tool accuracy.
• Consistent hotspot detection by both AI and manual methods, indicating clinical applicability for proliferation assessment.
• AI processing time reduced by approximately 95%, demonstrating significant efficiency gains in skeletal marrow and organ quantification.

Conclusions:

• The AI-driven segmentation platform is technically validated for FLT-PET/CT in AML and HCT, enabling separate quantification of cortical bone and trabecular marrow.
• The automated approach shows high agreement, excellent reproducibility, and substantial efficiency improvements, making it suitable for clinical workflows.
• This scalable framework facilitates future studies correlating FLT-based bone marrow metrics with clinical outcomes in AML patients.