Amandine Crombé1,2,3, Xavier Buy1, Fei Han4, Solenn Toupin5, Michèle Kind1. 1. Department of Oncologic Imaging, Institut Bergonié, Comprehensive Cancer Center of Nouvelle-Aquitaine, Bordeaux, France. 2. Bordeaux University, Bordeaux, France. 3. Modelisation in Oncology (MOnc) Team, INRIA Bordeaux-Sud-Ouest, CNRS UMR 5251, Talence, France. 4. Siemens Medical Solutions USA, Los Angeles, California, USA. 5. Siemens Healthcare France, Saint-Denis, France.
Abstract
BACKGROUND: Magnetic resonance imaging (MRI)-based radiomics features (RFs) quantify tumors radiological phenotypes but are sensitive to postprocessing parameters, including the intensity harmonization technique (IHT), while mappings enable objective quantitative assessment. PURPOSE: To investigate whether T2 mapping could improve repeatability, reproducibility, and performances of radiomics compared to conventional T2-weighted imaging (T2WI). STUDY TYPE: Prospective. SUBJECTS: Twenty-six healthy adults. FIELD STRENGTH/SEQUENCE: Respiratory-trigged radial turbo spin echo (TSE) multiecho T2 mapping (prototype) and conventional TSE T2WI of the abdomen were acquired twice at 1.5 T. ASSESSMENT: T2 maps were reconstructed using a two-parameter exponential fitting model. Volumes-of-interest (VOIs) were manually drawn in six tissues: liver, kidney, pancreas, muscle, bone, and spleen. After co-registration, conventional T2WIs were processed with two IHTs (standardization [std] and histogram-matching [HM]) resulting in four paired input image types: initial T2WI, T2WIstd , T2WIHM , and T2-map. VOIs were propagated to extract 45 RFs from MRI-1 and MRI-2 of each image type (LIFEx, v5.10). STATISTICAL TESTS: Influence of the input data type on RF values was evaluated with analysis of variance. RFs test-retest repeatability and reproducibility over multiple segmentations were evaluated with intra-class correlation coefficient (ICC). Correlations between k-means clusters and the six tissues depending on the RFs dataset were investigated with adjusted-Rand-index (ARI). RESULTS: About 41 of 45 (91.1%) RFs were significantly influenced by the input image type (P values < 0.05), which was the most influential factor on repeatability of RFs (P-value < 0.05). Repeatability ICCs from T2-map displayed intermediate values between the initial T2WI (range: 0.407-0.736) and the T2WIHM (range: 0.724-0.817). The number of RFs with interobserver and intraobserver reproducibility ICCs ≥ 0.90 was 37/45 (82.2%) for T2WIHM , 33/45 (73.3%) for T2WIstd , 31/45 (68.9%) for T2 map, and 25/45 (55.6%) for the initial T2WI. T2 map provided the best tissue discrimination (ARI = 0.414 vs. 0.157 with T2WIHM ). DATA CONCLUSION: T2 mapping provided RFs with moderate to substantial repeatability and reproducibility ICCs, along with the most preserved discriminative information. LEVEL OF EVIDENCE: 1 TECHNICAL EFFICACY: 1.
BACKGROUND: Magnetic resonance imaging (MRI)-based radiomics features (RFs) quantify tumors radiological phenotypes but are sensitive to postprocessing parameters, including the intensity harmonization technique (IHT), while mappings enable objective quantitative assessment. PURPOSE: To investigate whether T2 mapping could improve repeatability, reproducibility, and performances of radiomics compared to conventional T2-weighted imaging (T2WI). STUDY TYPE: Prospective. SUBJECTS: Twenty-six healthy adults. FIELD STRENGTH/SEQUENCE: Respiratory-trigged radial turbo spin echo (TSE) multiecho T2 mapping (prototype) and conventional TSE T2WI of the abdomen were acquired twice at 1.5 T. ASSESSMENT: T2 maps were reconstructed using a two-parameter exponential fitting model. Volumes-of-interest (VOIs) were manually drawn in six tissues: liver, kidney, pancreas, muscle, bone, and spleen. After co-registration, conventional T2WIs were processed with two IHTs (standardization [std] and histogram-matching [HM]) resulting in four paired input image types: initial T2WI, T2WIstd , T2WIHM , and T2-map. VOIs were propagated to extract 45 RFs from MRI-1 and MRI-2 of each image type (LIFEx, v5.10). STATISTICAL TESTS: Influence of the input data type on RF values was evaluated with analysis of variance. RFs test-retest repeatability and reproducibility over multiple segmentations were evaluated with intra-class correlation coefficient (ICC). Correlations between k-means clusters and the six tissues depending on the RFs dataset were investigated with adjusted-Rand-index (ARI). RESULTS: About 41 of 45 (91.1%) RFs were significantly influenced by the input image type (P values < 0.05), which was the most influential factor on repeatability of RFs (P-value < 0.05). Repeatability ICCs from T2-map displayed intermediate values between the initial T2WI (range: 0.407-0.736) and the T2WIHM (range: 0.724-0.817). The number of RFs with interobserver and intraobserver reproducibility ICCs ≥ 0.90 was 37/45 (82.2%) for T2WIHM , 33/45 (73.3%) for T2WIstd , 31/45 (68.9%) for T2 map, and 25/45 (55.6%) for the initial T2WI. T2 map provided the best tissue discrimination (ARI = 0.414 vs. 0.157 with T2WIHM ). DATA CONCLUSION: T2 mapping provided RFs with moderate to substantial repeatability and reproducibility ICCs, along with the most preserved discriminative information. LEVEL OF EVIDENCE: 1 TECHNICAL EFFICACY: 1.