Dataset Information

Multicenter Study on COVID-19 Lung Computed Tomography Segmentation with varying Glass Ground Opacities using Unseen Deep Learning Artificial Intelligence Paradigms: COVLIAS 1.0 Validation.

ABSTRACT: Variations in COVID-19 lesions such as glass ground opacities (GGO), consolidations, and crazy paving can compromise the ability of solo-deep learning (SDL) or hybrid-deep learning (HDL) artificial intelligence (AI) models in predicting automated COVID-19 lung segmentation in Computed Tomography (CT) from unseen data leading to poor clinical manifestations. As the first study of its kind, "COVLIAS 1.0-Unseen" proves two hypotheses, (i) contrast adjustment is vital for AI, and (ii) HDL is superior to SDL. In a multicenter study, 10,000 CT slices were collected from 72 Italian (ITA) patients with low-GGO, and 80 Croatian (CRO) patients with high-GGO. Hounsfield Units (HU) were automatically adjusted to train the AI models and predict from test data, leading to four combinations-two Unseen sets: (i) train-CRO:test-ITA, (ii) train-ITA:test-CRO, and two Seen sets: (iii) train-CRO:test-CRO, (iv) train-ITA:test-ITA. COVILAS used three SDL models: PSPNet, SegNet, UNet and six HDL models: VGG-PSPNet, VGG-SegNet, VGG-UNet, ResNet-PSPNet, ResNet-SegNet, and ResNet-UNet. Two trained, blinded senior radiologists conducted ground truth annotations. Five types of performance metrics were used to validate COVLIAS 1.0-Unseen which was further benchmarked against MedSeg, an open-source web-based system. After HU adjustment for DS and JI, HDL (Unseen AI) > SDL (Unseen AI) by 4% and 5%, respectively. For CC, HDL (Unseen AI) > SDL (Unseen AI) by 6%. The COVLIAS-MedSeg difference was < 5%, meeting regulatory guidelines.Unseen AI was successfully demonstrated using automated HU adjustment. HDL was found to be superior to SDL.

SUBMITTER: Suri JS

PROVIDER: S-EPMC9392994 | biostudies-literature | 2022 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Multicenter Study on COVID-19 Lung Computed Tomography Segmentation with varying Glass Ground Opacities using Unseen Deep Learning Artificial Intelligence Paradigms: COVLIAS 1.0 Validation.

Suri Jasjit S JS Agarwal Sushant S Saba Luca L Chabert Gian Luca GL Carriero Alessandro A Paschè Alessio A Danna Pietro P Mehmedović Armin A Faa Gavino G Jujaray Tanay T Singh Inder M IM Khanna Narendra N NN Laird John R JR Sfikakis Petros P PP Agarwal Vikas V Teji Jagjit S JS R Yadav Rajanikant R Nagy Ferenc F Kincses Zsigmond Tamás ZT Ruzsa Zoltan Z Viskovic Klaudija K Kalra Mannudeep K MK

Journal of medical systems 20220821 10

Variations in COVID-19 lesions such as glass ground opacities (GGO), consolidations, and crazy paving can compromise the ability of solo-deep learning (SDL) or hybrid-deep learning (HDL) artificial intelligence (AI) models in predicting automated COVID-19 lung segmentation in Computed Tomography (CT) from unseen data leading to poor clinical manifestations. As the first study of its kind, "COVLIAS 1.0-Unseen" proves two hypotheses, (i) contrast adjustment is vital for AI, and (ii) HDL is superio ...[more]

PMID: 35988110

Similar Datasets

Project description:Objective: Pure ground-glass opacity (GGO) nodules have been detected with increasing frequency using computed tomography (CT). We performed a retrospective study to clarify whether lung cancer patient prognoses correlated with pure GGO nodules. We also analyzed the clinical characters of patients with pure GGO nodules to provide diagnostic guidance on lung cancer identification and treatment of patients in clinical practice. Methods: We enrolled 39 of 1422 patients with pure GGO nodules who accepted surgical treatment of the lung cancer nodules, and reviewed materials from 404 patients to verify our conclusions. To discover which factors were prognostically significant, we used the Kaplan-Meier method to estimate the overall survival (OS) and progression-free survival (PFS) curves. Age, gender, smoking history, histology, tumor size, and stage were the factors examined in our study. We also performed subgroup and matching group analyses to clarify the correlation between the presence of pure GGO nodules and prognoses. Results: Pure GGO nodules were associated with non-smoking females that had adenocarcinoma. The prognoses of patients in the pure GGO nodule group was better than those in the non-pure GGO nodule group (p = 0.046). Age, grade, and stage (including tumor size and lymph node metastases) were had prognostic significance. In the matching group stage assessments, although patient prognoses were not significantly different among patients of the GGO group compared with thoses of the other group in long-term, while in the short term, patients with pure GGO nodules had longer PFS. Non-smoking female patients with lung cancer were more likely to have adenocarcinoma. Conclusions: As a subgroup of GGO nodules, pure GGO nodules predict a better prognosis in all lung cancer patients. Wheras our study showed that lung patients with pure GGO nodules in similar stages were not significantly different in long-term prognoses, in the short term; patients with pure GGO nodules had longer PFS.

Project description:PURPOSE:Intraoperative localization and resection of ill-defined pulmonary ground-glass opacities (GGOs) during minimally invasive pulmonary resection is technically challenging. Current preoperative techniques to facilitate localization of GGOs include microcoil and hook wire placement, both of which have logistic limitations, carry safety concerns, and do not help with margin assessment. In this clinical trial, we explored an alternative method involving near-infrared molecular imaging with a folate receptor-targeted agent, OTL38, to improve localization of GGOs and confirmation of resection margins. METHODS:In a human trial, 20 subjects with pulmonary GGOs who were eligible for video-assisted thoracoscopic surgery (VATS) resection received 0.025 mg/kg of OTL38 before the resection. The primary objectives were to (1) determine whether use of OTL38 allows safe localization of GGOs and assessment of margins during VATS and (2) determine patient, radiographic, and histopathologic variables that predict the amount of fluorescence during near-infrared imaging. RESULTS:We observed no toxicity. Of the 21 GGOs, 20 accumulated OTL38 and displayed fluorescence upon in situ or back table evaluation. Intraoperatively, near-infrared imaging localized 15 of 21 lesions whereas VATS alone localized 10 of 21 (p = 0.05). The addition of molecular imaging affected care of nine of 21 subjects by improving intraoperative localization (n = 6) and identifying close margins (n = 3). This approach was most effective for subpleural lesions measuring less than 2 cm. For lesions deeper than 1.5 cm from the pleural surface, intraoperative localization using fluorescent feedback was limited. CONCLUSIONS:This approach provides a safe alternative for intraoperative localization of small, peripherally located pulmonary lesions. In contrast to alternative localization techniques, use of OTL38 also allows confirmation of adequate margins. Future studies will compare this approach to alternative localization techniques in a clinical trial.

Dataset Information

Multicenter Study on COVID-19 Lung Computed Tomography Segmentation with varying Glass Ground Opacities using Unseen Deep Learning Artificial Intelligence Paradigms: COVLIAS 1.0 Validation.

Publications

Multicenter Study on COVID-19 Lung Computed Tomography Segmentation with varying Glass Ground Opacities using Unseen Deep Learning Artificial Intelligence Paradigms: COVLIAS 1.0 Validation.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets