Semi-automatic quantitative analysis of the pelvic bony structures on apparent diffusion coefficient maps based on...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Apparent diffusion coefficient (ADC) maps offer quantitative tissue information but require precise normal tissue values for accurate interpretation.
• Distinguishing normal from abnormal pelvic bony structures using ADC is challenging without established reference ranges.

Purpose of the Study:

• To develop a deep learning-based convolutional neural network (CNN) for segmenting pelvic bony structures.
• To establish reference ranges for ADC parameters in normal pelvic bony structures.

Main Methods:

• Retrospective analysis of 767 prostate cancer patients' quantitative ADC data.
• Development of a CNN for segmenting 8 pelvic bony structures in a subset of 288 patients.
• Automated segmentation and ADC reference range establishment (95% CI) using data from 405 non-treated and 74 treated patients.

Main Results:

• CNN achieved high segmentation accuracy (Dice scores 0.90-0.95) for pelvic bones.
• No significant differences in segmentation accuracy were observed across different scanners or treatment groups.
• Established preliminary ADC reference ranges (95% CI) for lumbar vertebra, sacrococcyx, ilium, acetabulum, femoral head, femoral neck, ischium, and pubis.

Conclusions:

• Preliminary reference ranges for ADC values in normal pelvic bony structures were established.
• Image acquisition parameters were found to influence ADC values, highlighting their importance in quantitative analysis.