Table 1.
Shape features found significant in publications.
| Shape feature | Author article | Tumor localization | Clinical outcome |
|---|---|---|---|
| Compactness | Aerts2, Naturecomms, 2014 | Lung cancers, Head and neck cancers | CI of 0.65 (NSCLC) and 0.69 (HNSCC) for survival prediction, with 3 other features (Statistics total energy, GLRL GLN, Wavelet HLH GLN) |
| He8, IOP Science, 2014 | Lung lesions (LIDC-IDRI) | 2 features (with average gray value) had CI between computer scores and the reader scores of 0.789 ± 0.014 for nodule subtlety/automatic segmentation | |
| Wang6, IEEE, 2016 | Lung lesions (LIDC-IDRI) | Prediction of malignant lung tumor (accuracy: 86% TS, 76% VS) from 15 random forest selected features, 3 of which were shape-based | |
| Pena7, Academic Radiology, 2017 | Lung lesions | Prediction of malignant lung tumor with an AUC of 0.92 ± 0.05 (P < 0.0001), with 2 shape features included in a signature of 4 features | |
| Bogowicz10, IJROBP, 2017 | Head and neck cancers | Nine features predicted HPV status, including 2 shape features, with AUC = 0.66 | |
| Huynh9, PLOS ONE, 2017 | Early stage NSCLC (post SBRT) | 7 AIP features were associated with distant metastases, 3 of which were shape-based, with CI = 0.648 | |
| Spherical Disproportion | Wang6, IEEE, 2016 | Lung lesions (LIDC-IDRI) | Prediction of malignant lung tumor (accuracy: 86% TS, 76% VS) from 15 random forest selected features, 3 of which were shape-based |
| Bogowicz10, IJROBP, 2017 | Head and neck cancers | Nine features predicted HPV status, including 2 shape features with AUC = 0.66 | |
| Huynh9, PLOS ONE, 2017 | Early stage NSCLC (post SBRT) | 7 AIP features were associated with distant metastases, 3 of which were shape-based, with CI = 0.648 | |
| Sphericity | Huynh9, PLOS ONE, 2017 | Early stage NSCLC (post SBRT) | 7 AIP features were associated with distant metastases, 3 of which were shape-based, with CI = 0.648 |
| Song11, IASLC, 2017 | LADC | 3 features, one of which was shape-based, were predictors of > 5% micropapillary component in LADCs with AUC = 0.61 | |
| Surface-To- Volume | Wang6, IEEE, 2016 | Lung lesions (LIDC-IDRI) | Prediction of malignant lung tumor (accuracy 86% TS, 76% VS) from 15 Random Forest selected features, 3 of which were shape-based |
| Surface Area | Chaddad28, Oncotarget, 2017 | NSCLC (TCIA) | Surface area was correlated with the survival time of patients with large cell carcinoma, T2, N0 and Stage I tumors with p < 0.05 |
| S1 (Max. Thickness of The Lesion Skeleton) in 2d | Pena7, Academic Radiology, 2017 | Lung lesions | Prediction of malignant lung tumor AUC = 0.92 ± 0.05 (P < 0.0001), with 2 shape features included in a signature of 4 |
NSCLC: non-small cell lung cancer, HNSCC: head and neck squamous cell carcinoma, CI: concordance index, GLRL: gray level run length, GLN: gray level non-uniformity, LIDC-IDRI: Lung Image Database Consortium, SBRT: stereotactic body radiotherapy, AIP: average intensity projection, LADC: Lung adenocarcinoma, TS: Training Set, VS: Validation Set, TCIA: The Cancer Imaging Archive, AUC: area under the curve, HPV: Human Papilloma Virus.