Computed tomography (CT)-derived radiomic features differentiate prevascular mediastinum masses as thymic neoplasms vers
- PDF / 1,937,797 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 87 Downloads / 155 Views
COMPUTED TOMOGRAPHY
Computed tomography (CT)‑derived radiomic features differentiate prevascular mediastinum masses as thymic neoplasms versus lymphomas Margarita Kirienko1 · Gaia Ninatti1 · Luca Cozzi1,2 · Emanuele Voulaz1,3 · Nicolò Gennaro4 · Isabella Barajon1 · Francesca Ricci5 · Carmelo Carlo‑Stella1,5 · Paolo Zucali5 · Martina Sollini1,6 · Luca Balzarini7 · Arturo Chiti1,6 Received: 11 December 2019 / Accepted: 30 March 2020 © Italian Society of Medical Radiology 2020
Abstract Objectives We aimed to assess the ability of radiomics, applied to not-enhanced computed tomography (CT), to differentiate mediastinal masses as thymic neoplasms vs lymphomas. Methods The present study was an observational retrospective trial. Inclusion criteria were pathology-proven thymic neoplasia or lymphoma with mediastinal localization, availability of CT. Exclusion criteria were age 0.2) Comparison between AUC α vs AUC 2B (variables): not significant (p>>0.5) AUC – area under the curve; LDA – linear discriminant analysis.
13
p=0.05* p=0.15 p>0.2 p>>0.5
Model 3
p>0.2
p>0.2 p=0.5
La radiologia medica
over time. Similarly, also the reproducibility of the results regardless image processing settings is an essential condition for generalizability. Our study is the first to provide the differential diagnosis in solid prevascular mediastinal masses, with independent data validation. Interestingly, the CT-based radiomic model was comparable to the clinical and radiomic + clinical ones. This could be particularly relevant in daily clinical practice. In fact, when signs and symptoms are typical, the clinical diagnosis of a mediastinal mass presents few difficulties. On the contrary, in asymptomatic patients
(42% in our series; clinical cases are presented in Fig. 1), or with atypical clinical presentation or in the absence of suggestive findings at imaging, radiomics might be helpful to reach non-invasively a diagnosis (Fig. 2). Previous radiomics studies focused on mediastinal lesions, which aimed at differentiating solid mediastinal masses from cysts [10] or at characterizing thymic neoplasms [11–14], provided uncertain results. In fact, these studies were affected by considerable methodological drawbacks including small sample size, suboptimal statistical
Fig. 2 Schematic representation of the study workflow and summary of the results: 1—patient selection, clinical and laboratory data collection from the institutional database; 2—non-contrast-enhanced computed tomography images retrieval and qualitative evaluation; 3—manual segmentation using two different software; 4—two dif-
ferent resampling settings; 5—radiomic feature calculation from the image datasets; 6—both clinical data and radiomic features tested as diagnostic predictors; 7—radiomic, clinical and radiomic + clinical models’ performance comparison; 8—study results significance. VOI—volume of interest
13
approach or reference standard, number of features tested heavily unbalanced compared to the number of patients, and the lack of any type
Data Loading...
