MR-based artificial intelligence model to assess response to therapy in locally advanced rectal cancer. Issue 118 (September 2019)

• Record Type: Journal Article
• Authors: Ferrari, R.; Mancini-Terracciano, C.; Voena, C.; Rengo, M.; Zerunian, M.; Ciardiello, A.; Grasso, S.; Mare', V.; Paramatti, R.; Russomando, A.; Santacesaria, R.; Satta, A.; Solfaroli Camillocci, E.; Faccini, R.; Laghi, A.
• Is Part Of: European journal of radiology. Issue 118(2019)
• Page Start: 1
• Page End: 9
• Publication Date: 2019-09
• Subjects: T2w T2-weighted -- AI artificial Intelligence -- CRT chemo-radiotherapy -- RF random forest -- CR complete responder -- PR partial responder -- NR non-responder -- LARC locally-advanced rectal cancer -- TME total mesorectal excision; Rectal cancer -- Magnetic resonance imaging -- Texture analysis -- Artificial intelligence -- Neoadjuvant chemoradiotherapy; Medical radiology -- Periodicals; Radiology -- Periodicals; Radiologie médicale -- Périodiques; Medical radiology; Periodicals; 616.075705
• DOI: 10.1016/j.ejrad.2019.06.013
• Languages: English
• ISSNs: 0720-048X

Abstract

Highlights: Artificial intelligence is used for assessing response to therapy in rectal cancer. Textural features extraction from high resolution 3 T MR images. Artificial intelligence model helps personalize therapeutic strategy. Decisions curves analysis confirm clinical utility. Abstract: Purpose: To develop and validate an Artificial Intelligence (AI) model based on texture analysis of high-resolution T2 weighted MR images able 1) to predict pathologic Complete Response (CR) and 2) to identify non-responders (NR) among patients with locally-advanced rectal cancer (LARC) after receiving neoadjuvant chemoradiotherapy (CRT). Method: Fifty-five consecutive patients with LARC were retrospectively enrolled in this study. Patients underwent 3 T Magnetic Resonance Imaging (MRI) acquiring T2-weighted images before, during and after CRT. All patients underwent complete surgical resection and histopathology was the gold standard. Textural features were automatically extracted using an open-source software. A sub-set of statistically significant textural features was selected and two AI models were built by training a Random Forest (RF) classifier on 28 patients (training cohort). Model performances were estimated on 27 patients (validation cohort) using a ROC curve and a decision curve analysis. Results: Sixteen of 55 patients achieved CR. The AI model for CR classification showed good discrimination power with mean area under the receiver operating curve (AUC) of 0.86 (95% CI: 0.70, 0.94) in the validation cohort. The discriminatory power for the NR classification showed a mean AUC of 0.83 (95% CI: 0.71, 0.92). Decision curve analysis confirmed higher net patient benefit when using AI models compared to standard-of-care. Conclusions: AI models based on textural features of MR images of patients with LARC may help to identify patients who will show CR at the end of treatment and those who will not respond to therapy (NR) at an early stage of the treatment.