Cargando…
Magnetic-Resonance-Imaging Texture Analysis Predicts Early Progression in Rectal Cancer Patients Undergoing Neoadjuvant Chemoradiation
BACKGROUND: We hypothesized that texture analysis (TA) from the preoperative MRI can predict early disease progression (ePD), defined as the percentage of patients who relapsed or showed distant metastasis within three months from the radical surgery, in patients with locally advanced rectal cancer...
Autores principales: | , , , , , , , , , , , , , |
---|---|
Formato: | Online Artículo Texto |
Lenguaje: | English |
Publicado: |
Hindawi
2019
|
Materias: | |
Acceso en línea: | https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6360039/ https://www.ncbi.nlm.nih.gov/pubmed/30847005 http://dx.doi.org/10.1155/2019/8505798 |
Sumario: | BACKGROUND: We hypothesized that texture analysis (TA) from the preoperative MRI can predict early disease progression (ePD), defined as the percentage of patients who relapsed or showed distant metastasis within three months from the radical surgery, in patients with locally advanced rectal cancer (LARC, stage II and III, AJCC) undergoing neoadjuvant chemoradiotherapy (C-RT). METHODS: This retrospective monoinstitutional cohort study included 49 consecutive patients in total with a newly diagnosed rectal cancer. All the patients underwent baseline abdominal MRI and CT scan of the chest and abdomen to exclude distant metastasis before C-RT. Texture parameters were extracted from MRI performed before C-RT (T1, DWI, and ADC sequences) using LifeX Software, a dedicated software for extracting texture parameters from radiological imaging. We divided the cohort in a training set of 34 patients and a validation set of 15 patients, and we tested the data sets for homogeneity, considering the clinical variables. Then we performed univariate and multivariate analysis, and a ROC curve was also generated. RESULTS: Thirteen patients (26.5%) showed an ePD, three of whom with lung metastases and ten with liver relapse. The model was validated based on the prediction accuracy calculated in a previously unseen set of 15 patients. The prediction accuracy of the generated model was 82% (AUC = 0.853) in the training and 80% (AUC = 0.833) in the validation cohort. The only significant features at multivariate analysis was DWI GLCM Correlation (OR: 0.239, p < 0.001). CONCLUSION: Our results suggest that TA could be useful to identify patients that may develop early progression. |
---|