FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3048991087> ?p ?o ?g. }

• W3048991087 endingPage "507" @default.
• W3048991087 startingPage "507" @default.
• W3048991087 abstract "507 Objectives: Detection of pathological mediastinal nodes in non-small cell lung cancer (NSCLC) patients from 18F-FDG PET/CT images is a time-consuming task performed by Nuclear Medicine physicians (NMP) / radiologists that strongly impacts prognosis and patient management. Yet, current detection performance is fair (Specificity and Sensitivity of 0.93 and 0.65 respectively (1)). We investigated whether an automated approach based on handcrafted and deep radiomic features could be used to assist physicians with this task. Methods: A cohort of 97 NSCLC patients who underwent 18F-FDG PET/CT scans was studied (Gemini TF; Philips Medical Systems, Best, the Netherlands). Coordinates of pathological mediastinal nodes were marked by a dual-board certified radiologist (15y experience in thoracic imaging), yielding 151 pathological nodes. Two methods were used to learn features relevant to the detection of these nodes. For both methods, the scans were broken down into smaller overlapping cubes for easier analysis. Method 1 extracted 642 handcrafted radiomic features (321 from PET and from CT) for each cube using PyRadiomics, including first order, second order and wavelet-image features. Method 2 input both PET and CT cubes simultaneously into a two-branch 3D Convolutional Neural Network (CNN), from which 512 features comprising the final layer were extracted. The CNN had been pre-trained on a self-supervised learning task. Features from methods 1 and 2 were used separately and in combination to train a support vector machine (SVM) to decide whether each cube included a pathological node. The resulting detection algorithms were assessed using an independent test set of 30 patients with the physician as a ground truth. The test set was then independently evaluated by a second physician (5y experience in thoracic imaging). Sensitivity and positive predictive value (PPV) of this second physician with respect to the first were compared to sensitivity and PPV of the algorithm with respect to the first physician. Results: Results are shown in Table 1, with 95% confidence intervals obtained by bootstrapping. They show no significant difference in performance between the handcrafted and CNN feature-based models. There was no advantage in combining the handcrafted and CNN features. The sensitivity and PPV using the combined features were 0.66 and 0.54 respectively, comparable to the values obtained by comparing one physician with another (sensitivity and PPV of 0.65 and 0.53 respectively). Conclusions: SVM models involving either deep or handcrafted radiomic features could detect pathological mediastinal nodes in NSCLC patients with performance similar to those of NMP/radiologists. Further validation of the models against biopsies is warranted." @default.
• W3048991087 created "2020-08-21" @default.
• W3048991087 creator A5007085009 @default.
• W3048991087 creator A5059978336 @default.
• W3048991087 creator A5071620424 @default.
• W3048991087 creator A5076753728 @default.
• W3048991087 creator A5083883351 @default.
• W3048991087 creator A5088633704 @default.
• W3048991087 date "2020-05-01" @default.
• W3048991087 modified "2023-10-03" @default.
• W3048991087 title "Automated detection of mediastinal cancer nodes from 18F-FDG PET/CT scans of NSCLC patients using radiomic and CNN features" @default.
• W3048991087 hasPublicationYear "2020" @default.
• W3048991087 type Work @default.
• W3048991087 sameAs 3048991087 @default.
• W3048991087 citedByCount "0" @default.
• W3048991087 crossrefType "journal-article" @default.
• W3048991087 hasAuthorship W3048991087A5007085009 @default.
• W3048991087 hasAuthorship W3048991087A5059978336 @default.
• W3048991087 hasAuthorship W3048991087A5071620424 @default.
• W3048991087 hasAuthorship W3048991087A5076753728 @default.
• W3048991087 hasAuthorship W3048991087A5083883351 @default.
• W3048991087 hasAuthorship W3048991087A5088633704 @default.
• W3048991087 hasConcept C108583219 @default.
• W3048991087 hasConcept C121608353 @default.
• W3048991087 hasConcept C126322002 @default.
• W3048991087 hasConcept C126838900 @default.
• W3048991087 hasConcept C142724271 @default.
• W3048991087 hasConcept C153180895 @default.
• W3048991087 hasConcept C154945302 @default.
• W3048991087 hasConcept C2776256026 @default.
• W3048991087 hasConcept C2779013556 @default.
• W3048991087 hasConcept C2780379385 @default.
• W3048991087 hasConcept C2989005 @default.
• W3048991087 hasConcept C41008148 @default.
• W3048991087 hasConcept C71924100 @default.
• W3048991087 hasConcept C81363708 @default.
• W3048991087 hasConceptScore W3048991087C108583219 @default.
• W3048991087 hasConceptScore W3048991087C121608353 @default.
• W3048991087 hasConceptScore W3048991087C126322002 @default.
• W3048991087 hasConceptScore W3048991087C126838900 @default.
• W3048991087 hasConceptScore W3048991087C142724271 @default.
• W3048991087 hasConceptScore W3048991087C153180895 @default.
• W3048991087 hasConceptScore W3048991087C154945302 @default.
• W3048991087 hasConceptScore W3048991087C2776256026 @default.
• W3048991087 hasConceptScore W3048991087C2779013556 @default.
• W3048991087 hasConceptScore W3048991087C2780379385 @default.
• W3048991087 hasConceptScore W3048991087C2989005 @default.
• W3048991087 hasConceptScore W3048991087C41008148 @default.
• W3048991087 hasConceptScore W3048991087C71924100 @default.
• W3048991087 hasConceptScore W3048991087C81363708 @default.
• W3048991087 hasLocation W30489910871 @default.
• W3048991087 hasOpenAccess W3048991087 @default.
• W3048991087 hasPrimaryLocation W30489910871 @default.
• W3048991087 hasRelatedWork W1179679491 @default.
• W3048991087 hasRelatedWork W1975655647 @default.
• W3048991087 hasRelatedWork W1994717090 @default.
• W3048991087 hasRelatedWork W2020167497 @default.
• W3048991087 hasRelatedWork W2059939049 @default.
• W3048991087 hasRelatedWork W2135285836 @default.
• W3048991087 hasRelatedWork W2800038717 @default.
• W3048991087 hasRelatedWork W2899393783 @default.
• W3048991087 hasRelatedWork W2915885988 @default.
• W3048991087 hasRelatedWork W2921137012 @default.
• W3048991087 hasRelatedWork W2948770679 @default.
• W3048991087 hasRelatedWork W3003475598 @default.
• W3048991087 hasRelatedWork W3035869245 @default.
• W3048991087 hasRelatedWork W3081846813 @default.
• W3048991087 hasRelatedWork W3116875353 @default.
• W3048991087 hasRelatedWork W3161152630 @default.
• W3048991087 hasRelatedWork W3195715374 @default.
• W3048991087 hasRelatedWork W3201539940 @default.
• W3048991087 hasRelatedWork W3205435018 @default.
• W3048991087 hasRelatedWork W3205948795 @default.
• W3048991087 hasVolume "61" @default.
• W3048991087 isParatext "false" @default.
• W3048991087 isRetracted "false" @default.
• W3048991087 magId "3048991087" @default.
• W3048991087 workType "article" @default.