Abstract
Rationale and Objectives The study aimed to identify a radiomic approach based on CT and or magnetic resonance (MR) features (shape and texture) that may help differentiate benign versus malignant pleural lesions, and to assess if the radiomic model may improve confidence and accuracy of radiologists with different subspecialty backgrounds. Materials and Methods Twenty-nine patients with pleural lesions studied on both contrast-enhanced CT and MR imaging were reviewed retrospectively. Three texture and three shape features were extracted. Combinations of features were used to generate logistic regression models using histopathology as outcome. Two thoracic and two abdominal radiologists evaluated their degree of confidence in malignancy. Diagnostic accuracy of radiologists was determined using contingency tables. Cohen's kappa coefficient was used to assess inter-reader agreement. Using optimal threshold criteria, sensitivity, specificity, and accuracy of each feature and combination of features were obtained and compared to the accuracy and confidence of radiologists. Results The CT model that best discriminated malignant from benign lesions revealed an AUCCT = 0.92 ± 0.05 (P < 0.0001). The most discriminative MR model showed an AUCMR = 0.87 ± 0.09 (P < 0.0001). The CT model was compared to the diagnostic confidence of all radiologists and the model outperformed both abdominal radiologists (P < 0.002), whereas the top discriminative MR model outperformed one of the abdominal radiologists (P = 0.02). The most discriminative MR model was more accurate than one abdominal (P = 0.04) and one thoracic radiologist (P = 0.02). Conclusion Quantitative textural and shape analysis may help distinguish malignant from benign lesions. A radiomics-based approach may increase diagnostic confidence of abdominal radiologists on CT and MR and may potentially improve radiologists' accuracy in the assessment of pleural lesions characterized by MR.
Original language | English |
---|---|
Pages (from-to) | 1277-1287 |
Number of pages | 11 |
Journal | Academic Radiology |
Volume | 24 |
Issue number | 10 |
DOIs | |
Publication status | Published - Oct 2017 |
Externally published | Yes |
Bibliographical note
Publisher Copyright:© 2017 The Association of University Radiologists
ASJC Scopus Subject Areas
- Radiology Nuclear Medicine and imaging
PubMed: MeSH publication types
- Journal Article