Robust Radiomics feature quantification using semiautomatic volumetric segmentation.

Area of Science:

• Medical Imaging Analysis
• Radiomics
• Quantitative Imaging

Background:

• Advances in medical imaging allow tumor phenotype quantification.
• Radiomics converts images into data by extracting quantitative features.
• Tumor segmentation is a key challenge in Radiomics, with manual methods being time-consuming and variable.

Purpose of the Study:

• To investigate the robustness of a semi-automatic region growing volumetric segmentation algorithm in 3D-Slicer for quantitative imaging feature extraction.
• To compare the reproducibility and robustness of radiomic features extracted from 3D-Slicer segmentations versus manual delineations.

Main Methods:

• A semi-automatic region growing volumetric segmentation algorithm in 3D-Slicer was used.
• Fifty-six 3D-radiomic features were extracted from CT images of 20 lung cancer patients.
• Features were derived from 3D-tumor volumes segmented by observers using 3D-Slicer and compared to manual slice-by-slice delineations.

Main Results:

• Radiomic features from 3D-Slicer segmentations showed significantly higher reproducibility (ICC = 0.85 ± 0.15) than manual segmentations (ICC = 0.77 ± 0.17).
• Features extracted using 3D-Slicer were more robust, exhibiting a significantly smaller range across observers.
• 3D-Slicer segmentations demonstrated better overlap with manual contouring feature ranges.

Conclusions:

• 3D-Slicer segmented tumor volumes offer a superior alternative to manual delineation for radiomic feature quantification.
• The use of 3D-Slicer leads to more reproducible and robust imaging descriptors.
• 3D-Slicer is suitable for quantitative image feature extraction and data mining in large patient cohorts.