Abstract
505
Objectives: The calculation of textural features (TF) from PET images is gaining interest to complement conventional parameters such as SUVs, Metabolic Volumes (MV), and Total Lesion Glycolysis. Yet, depending on how they are calculated, TF values are often significantly correlated either with SUV or with MV, making them difficult to interpret. To clarify the sensitivity of TF to the actual spatial arrangement of metabolic activity, we investigated whether two tissue types with similar average SUV could be distinguished based on TF only.
Methods: 111 patients with breast cancer without any node involvement nor metastatic spread underwent a PET/CT 75 min± 9 after injection of 3-3.5 MBq/kg FDG on a Gemini TF PET/CT scanner. PET images (4 x 4 x 4 mm3 voxel size) were reconstructed using a BLOB-OS-TF algorithm. In each patient, 6 spherical VOI of 23 mL were drawn in the healthy liver, spleen, lung, fat, muscle and healthy contralateral breast. The mean SUV (SUVmean) and 6 TF (homogeneity, entropy, LRE, SRE, LGZE and HGZE) were measured in each VOI using fixed bin size for quantization. The ability of TF to distinguish between different tissue types was tested using Wilcoxon test for the entire cohort and for pairs of VOI with a difference of SUVmean lower than 0.01 SUV, with four comparisons: liver and spleen tissues, breast and fat tissues, breast and lung tissues and breast and muscle tissues.
Results: For the entire cohort, all indices (SUVmean and TF) were significantly different between all tissues, except Entropy between liver and spleen tissues and between muscle and breast tissues. When comparing VOIs with similar SUVmean, Homogeneity, Entropy, SRE and LRE could distinguish between liver and spleen tissues, and also between breast and fat, between breast and lung, and between breast and muscle tissues. For example, liver Entropy was 1.06 on average, but spleen Entropy tended to be systematically higher by 0.1, suggesting a more disorganized metabolic activity in spleen tissue than in liver tissue for VOIs with identical SUVmean. Breast Homogeneity tended to be systematically lower than fat tissue Homogeneity (0.90±0.05 in breast versus 0.93±0.05 in fat), than lung tissue Homogeneity (0.82±0.05 versus 0.87±0.04) and than muscle tissue Homogeneity (0.76±0.4 versus 0.81±0.04) when matching SUVmean in the VOI comparison. HGZE was significantly different between different tissues for all comparisons of VOIs with similar SUVmean except between fat and breast. LGZE could also distinguish between each pair of tissue type with matched SUVmean except for fat and breast and for muscle and breast.
Conclusion: When examining tissues with similar SUVs in VOIs of identical size, we found that TF only could identify the tissue type, demonstrating that TF measured in PET images is sensitive enough to reflect subtle difference in metabolic activity spatial organization. This suggests that in patients, TF can distinguish between a low density of highly metabolically active cells and a high density of moderately metabolically active cells that yield similar SUVmean. Research Support: ANR-11-IDEX-0003-02