Author of 1 Presentation
SS 4.2 - CT texture analysis of GI stromal tumours
Abstract
Purpose
To evaluate the association between radiomic biomarkers extracted from baseline CT imaging, mitotic count, tumor mutational profile and prognostic Miettinen classification.
Material and methods
This retrospective multicenter observational study includes 63 histologically proven gastrointestinal stromal tumors (GISTs). Each lesion was manually segmented; 37 texture features were extracted either on a single slice and on the entire tumor volume. Reference standards: pathological findings and Miettinen classification. Patients were dichotomized with mitotic count (≤5/50HPF vs >5/50HPF), mutational status (c-KIT mutation vs PDGFRα and wild-type), patients prognosis (good prognosis class: none, very low and low risk vs poor prognosis class: intermediate and high risk). Univariate analysis using the Mann-Whitney test and multivariate analysis were performed; a stepwise logistic regression model was developed to predict patient's prognosis using 70% of patients as the training set and the remaining 30% as the test set.
Results
Eight 3D features discriminated lesions with low or high mitotic count (best AUC 0.81, best sensitivity 86%, best specificity 93%). Six 3D parameters detected GISTs based on the mutational group (best AUC 0.77, best sensitivity 75%, best specificity 79%) and three parameters correlated with risk class (best AUC 0.76, best sensitivity 72%, best specificity 85%). To differentiate between GIST at lower or higher risk of recurrence, the regression model used 6 different features with AUC 0.78, sensitivity 65%, specificity 79%, VPN 71% and VPP 73% on the training set, and AUC 0.83, sensitivity 88% and specificity 75% on the test set.
Conclusion
A good correlation between radiomics features, disease aggressiveness, mutational profile and risk of recurrence was observed. Results are promising; validation on external datasets is necessary to confirm the role as imaging biomarker.
Video-on-demand
Presenter of 1 Presentation
SS 4.2 - CT texture analysis of GI stromal tumours
Abstract
Purpose
To evaluate the association between radiomic biomarkers extracted from baseline CT imaging, mitotic count, tumor mutational profile and prognostic Miettinen classification.
Material and methods
This retrospective multicenter observational study includes 63 histologically proven gastrointestinal stromal tumors (GISTs). Each lesion was manually segmented; 37 texture features were extracted either on a single slice and on the entire tumor volume. Reference standards: pathological findings and Miettinen classification. Patients were dichotomized with mitotic count (≤5/50HPF vs >5/50HPF), mutational status (c-KIT mutation vs PDGFRα and wild-type), patients prognosis (good prognosis class: none, very low and low risk vs poor prognosis class: intermediate and high risk). Univariate analysis using the Mann-Whitney test and multivariate analysis were performed; a stepwise logistic regression model was developed to predict patient's prognosis using 70% of patients as the training set and the remaining 30% as the test set.
Results
Eight 3D features discriminated lesions with low or high mitotic count (best AUC 0.81, best sensitivity 86%, best specificity 93%). Six 3D parameters detected GISTs based on the mutational group (best AUC 0.77, best sensitivity 75%, best specificity 79%) and three parameters correlated with risk class (best AUC 0.76, best sensitivity 72%, best specificity 85%). To differentiate between GIST at lower or higher risk of recurrence, the regression model used 6 different features with AUC 0.78, sensitivity 65%, specificity 79%, VPN 71% and VPP 73% on the training set, and AUC 0.83, sensitivity 88% and specificity 75% on the test set.
Conclusion
A good correlation between radiomics features, disease aggressiveness, mutational profile and risk of recurrence was observed. Results are promising; validation on external datasets is necessary to confirm the role as imaging biomarker.
Video-on-demand
Author of 1 Presentation
SS 4.2 - CT texture analysis of GI stromal tumours (ID 884)
Abstract
Purpose
To evaluate the association between radiomic biomarkers extracted from baseline CT imaging, mitotic count, tumor mutational profile and prognostic Miettinen classification.
Material and methods
This retrospective multicenter observational study includes 63 histologically proven gastrointestinal stromal tumors (GISTs). Each lesion was manually segmented; 37 texture features were extracted either on a single slice and on the entire tumor volume. Reference standards: pathological findings and Miettinen classification. Patients were dichotomized with mitotic count (≤5/50HPF vs >5/50HPF), mutational status (c-KIT mutation vs PDGFRα and wild-type), patients prognosis (good prognosis class: none, very low and low risk vs poor prognosis class: intermediate and high risk). Univariate analysis using the Mann-Whitney test and multivariate analysis were performed; a stepwise logistic regression model was developed to predict patient's prognosis using 70% of patients as the training set and the remaining 30% as the test set.
Results
Eight 3D features discriminated lesions with low or high mitotic count (best AUC 0.81, best sensitivity 86%, best specificity 93%). Six 3D parameters detected GISTs based on the mutational group (best AUC 0.77, best sensitivity 75%, best specificity 79%) and three parameters correlated with risk class (best AUC 0.76, best sensitivity 72%, best specificity 85%). To differentiate between GIST at lower or higher risk of recurrence, the regression model used 6 different features with AUC 0.78, sensitivity 65%, specificity 79%, VPN 71% and VPP 73% on the training set, and AUC 0.83, sensitivity 88% and specificity 75% on the test set.
Conclusion
A good correlation between radiomics features, disease aggressiveness, mutational profile and risk of recurrence was observed. Results are promising; validation on external datasets is necessary to confirm the role as imaging biomarker.
Video-on-demand
Presenter of 1 Presentation
SS 4.2 - CT texture analysis of GI stromal tumours (ID 884)
Abstract
Purpose
To evaluate the association between radiomic biomarkers extracted from baseline CT imaging, mitotic count, tumor mutational profile and prognostic Miettinen classification.
Material and methods
This retrospective multicenter observational study includes 63 histologically proven gastrointestinal stromal tumors (GISTs). Each lesion was manually segmented; 37 texture features were extracted either on a single slice and on the entire tumor volume. Reference standards: pathological findings and Miettinen classification. Patients were dichotomized with mitotic count (≤5/50HPF vs >5/50HPF), mutational status (c-KIT mutation vs PDGFRα and wild-type), patients prognosis (good prognosis class: none, very low and low risk vs poor prognosis class: intermediate and high risk). Univariate analysis using the Mann-Whitney test and multivariate analysis were performed; a stepwise logistic regression model was developed to predict patient's prognosis using 70% of patients as the training set and the remaining 30% as the test set.
Results
Eight 3D features discriminated lesions with low or high mitotic count (best AUC 0.81, best sensitivity 86%, best specificity 93%). Six 3D parameters detected GISTs based on the mutational group (best AUC 0.77, best sensitivity 75%, best specificity 79%) and three parameters correlated with risk class (best AUC 0.76, best sensitivity 72%, best specificity 85%). To differentiate between GIST at lower or higher risk of recurrence, the regression model used 6 different features with AUC 0.78, sensitivity 65%, specificity 79%, VPN 71% and VPP 73% on the training set, and AUC 0.83, sensitivity 88% and specificity 75% on the test set.
Conclusion
A good correlation between radiomics features, disease aggressiveness, mutational profile and risk of recurrence was observed. Results are promising; validation on external datasets is necessary to confirm the role as imaging biomarker.