1
|
Fanni SC, Romei C, Ferrando G, Volpi F, D’Amore CA, Bedini C, Ubbiali S, Valentino S, Neri E. Natural language processing to convert unstructured COVID-19 chest-CT reports into structured reports. Eur J Radiol Open 2023; 11:100512. [PMID: 37575311 PMCID: PMC10413059 DOI: 10.1016/j.ejro.2023.100512] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 05/27/2023] [Revised: 07/17/2023] [Accepted: 07/18/2023] [Indexed: 08/15/2023] Open
Abstract
Background Structured reporting has been demonstrated to increase report completeness and to reduce error rate, also enabling data mining of radiological reports. Still, structured reporting is perceived by radiologists as a fragmented reporting style, limiting their freedom of expression. Purpose A deep learning-based natural language processing method was developed to automatically convert unstructured COVID-19 chest CT reports into structured reports. Methods Two hundred-two COVID-19 chest CT were retrospectively reviewed by two experienced radiologists, who wrote for each exam a free-form text radiological report and coherently filled the template provided by the Italian Society of Medical and Interventional Radiology, used as ground-truth. A semi-supervised convolutional neural network was implemented to extract 62 categorical variables from the report. Two iterations were carried-out, the first without fine-tuning, the second one performing a fine-tuning. The performance was measured using the mean accuracy and the F1 mean score. An error analysis was performed to identify errors entirely attributable to incorrect processing of the model. Results The algorithm achieved a mean accuracy of 93.7% and an F1 score 93.8% in the first iteration. Most of the errors were exclusively attributable to wrong inference (46%). In the second iteration the model achieved for both parameters 95,8% and percentage of errors attributable to wrong inference decreased to 26%. Conclusions The convolutional neural network achieved an optimal performance in the automated conversion of free-form text into structured radiological reports, overcoming all the limitation attributed to structured reporting and finally paving the way for data mining of radiological report.
Collapse
Affiliation(s)
| | - Chiara Romei
- Department of Diagnostic Imaging, 2nd Radiology Unit, Pisa University-Hospital, Pisa, Italy
| | | | - Federica Volpi
- Department of Translational Research, Academic Radiology, University of Pisa, Pisa, Italy
| | - Caterina Aida D’Amore
- Department of Translational Research, Academic Radiology, University of Pisa, Pisa, Italy
| | | | - Sandro Ubbiali
- EBIT sr.l. Esaote Group, Via di Caciolle, Florence, Italy
| | | | - Emanuele Neri
- Department of Translational Research, Academic Radiology, University of Pisa, Pisa, Italy
| |
Collapse
|
2
|
Maciel C, Ferreira H, Djokovic D, Kyaw Tun J, Keckstein J, Rizzo S, Manganaro L. MRI of endometriosis in correlation with the #Enzian classification: applicability and structured report. Insights Imaging 2023; 14:120. [PMID: 37405519 DOI: 10.1186/s13244-023-01466-x] [Citation(s) in RCA: 4] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [Abstract] [Key Words] [Journal Information] [Subscribe] [Scholar Register] [Received: 02/06/2023] [Accepted: 06/14/2023] [Indexed: 07/06/2023] Open
Abstract
Endometriosis represents one of the most common causes of life-impacting chronic pelvic pain and female infertility. Magnetic resonance imaging (MRI) plays an increasing role in the diagnosis and mapping of endometriosis, while diagnostic laparoscopy currently tends to be reserved for the patients with negative imaging results. The #Enzian, published in 2021, proposes a new comprehensive classification system of endometriosis, combining a complete staging of deep infiltrative endometriosis with the evaluation of peritoneal/ovarian/tubal localizations and the presence of adenomyosis. This article addresses in detail the applicability of the #Enzian classification, primarily based on surgical findings, to the MRI evaluation of the endometriosis. Overall, there is a significant matching between MRI features and the #Enzian classification criteria, two different perspectives of endometriosis mapping, with different goals and levels of detail. The main discrepancy lies in the evaluation of tubo-ovarian condition, which is not fully assessable by MRI. Furthermore, as endometriosis is a complex disease, usually multifocal, that can present with a myriad of imaging findings, MRI reporting should be clear and well organized. The authors group, both radiologists and gynecologists, propose a structured MRI report of endometriosis in correlation with the #Enzian classification, merging the detailed anatomical and pre-operative information provided by the MRI with the benefits of a comprehensive classification system of endometriosis in the clinical practice and research field.Critical relevance statement This article addresses in detail the applicability of the #Enzian classification, primarily based on surgical findings, to the MRI evaluation of the endometriosis and proposes a #Enzian-based structured MRI report.
Collapse
Affiliation(s)
- Cristina Maciel
- Serviço de Radiologia, Centro Hospitalar Universitário São João, Porto, Portugal.
- Departamento de Medicina, Faculdade de Medicina, Universidade do Porto, Alameda Prof. Hernâni Monteiro, 4200-319, Porto, Portugal.
| | - Hélder Ferreira
- Serviço de Ginecologia, Centro Materno Infantil do Norte, Centro Hospitalar Universitário do Porto, Largo do Prof. Abel Salazar, 4099-001, Porto, Portugal
- Instituto de Ciências Biomédicas Abel Salazar, Universidade do Porto, Rua de Jorge Viterbo Ferreira, 228, 4050-313, Porto, Portugal
| | - Dusan Djokovic
- Maternidade Dr. Alfredo da Costa, Centro Hospitalar Universitário Lisboa Central (CHULC), Lisbon, Portugal
- Department of Obstetrics and Gynecology, NOVA Medical School - Faculdade de Ciências Médicas, NOVA University of Lisbon, Lisbon, Portugal
- Department of Obstetrics and Gynecology, Hospital CUF Descobertas, Lisbon, Portugal
| | - Jimmy Kyaw Tun
- Department of Interventional Radiology, The Royal London Hospital, Barts Health NHS Trust, London, E1 1BB, UK
| | - Jörg Keckstein
- Scientific Endometriosis Foundation (Stiftung Endometrioseforschung/SEF), Westerstede, Germany
- Endometriosis Clinic Dres. Keckstein, Villach, Austria
- University of Ulm, Ulm, Germany
| | - Stefania Rizzo
- Service of Radiology, Imaging Institute of Southern Switzerland, EOC, Via Tesserete 46, 6900, Lugano, Switzerland
- Faculty of Biomedical Sciences, Università Della Svizzera Italiana, Via G. Buffi 13, 6900, Lugano, Switzerland
| | - Lucia Manganaro
- Department of Radiological, Oncological and Pathological Sciences, Sapienza University of Rome, Rome, Italy
| |
Collapse
|
3
|
Malik RF, Hasanain A, Lafaro KJ, He J, Narang AK, Fishman EK, Zaheer A. Structured CT reporting of pancreatic ductal adenocarcinoma: impact on completeness of information and interdisciplinary communication for surgical planning. Abdom Radiol (NY) 2022; 47:704-714. [PMID: 34800162 DOI: 10.1007/s00261-021-03353-9] [Citation(s) in RCA: 3] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/22/2021] [Revised: 11/12/2021] [Accepted: 11/15/2021] [Indexed: 10/19/2022]
Abstract
PURPOSE With the rise in popularity of structured reports in radiology, we sought to evaluate whether free-text CT reports on pancreatic ductal adenocarcinoma (PDAC) staging at our institute met published guidelines and assess feedback of pancreatic surgeons comparing free-text and structured report styles with the same information content. METHODS We retrospectively evaluated 298 free-text preoperative CT reports from 2015 to 2017 for the inclusion of key tumor descriptors. Two surgeons independently evaluated 50 free-text reports followed by evaluation of the same reports in a structured format using a 7-question survey to assess the usefulness and ease of information extraction. Fisher's exact test and Chi-square test for independence were utilized for categorical responses and an independent samples t test for comparing mean ratings of report quality as rated on a 5-point Likert scale. RESULTS The most commonly included descriptors in free-text reports were tumor location (99%), liver lesions (97%), and suspicious lymph nodes (97%). The most commonly excluded descriptors were variant arterial anatomy and peritoneal/omental nodularity, which were present in only 23% and 42% of the reports, respectively. For vascular involvement, a mention of the presence or absence of perivascular disease with the main portal vein was most commonly included (87%). Both surgeons' rating of overall report quality was significantly higher for structured reports (p < 0.001). CONCLUSION Our results indicate that free-text reports may not include key descriptors for staging PDAC. Surgeons rated structured reports that presented the same information as free-text reports but in a template format superior for guiding clinical management, convenience of use, and overall report quality.
Collapse
|
4
|
Performance of imaging interpretation, intra- and inter-reader agreement for diagnosis of pelvic endometriosis: comparison between an abbreviated and full MRI protocol. Abdom Radiol (NY) 2021; 46:4025-4035. [PMID: 33772612 DOI: 10.1007/s00261-021-03052-5] [Citation(s) in RCA: 2] [Impact Index Per Article: 0.7] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/20/2020] [Revised: 03/03/2021] [Accepted: 03/09/2021] [Indexed: 12/29/2022]
Abstract
PURPOSE To compare the performance of imaging interpretation, intra- and inter-reader agreement between an abbreviated (aMRI) and full (fMRI) MRI protocol for diagnosis of pelvic endometriosis. METHODS Seventy consecutive fMRI exams performed under suspicion of pelvic endometriosis were selected. Four radiologists (Rd) (1-10 years experience) independently evaluated presence/absence of endometriosis at 9 anatomic sites (AS). The readers evaluated aMRI (coronal T2 TSE volumetric images and axial T1 GRE fat-sat without contrast, extracted from fMRI) and fMRI protocols randomly, with at least 4 weeks interval between readings. The degree of confidence for diagnosis at each AS was evaluated with a 1-3 Likert Scale (1: low; 3: high). Intra- and inter-reader agreement between protocols were evaluated by kappa statistics and took reading experience into account. The gold standard for assessing the performance of imaging interpretation (sensitivity, specificity and accuracy) used a consensus reading of two other Rd (> 15 years experience). RESULTS There was no significant difference in the accuracy of imaging interpretation between the abbreviated (0.83-0.86) and full (0.83-0.87) protocols (p = 0.15). Intra-reader agreement between protocols ranged from substantial to almost perfect (0.74-0.96). A substantial inter-reader agreement was found for both protocols for readers with similar levels of experience (0.67-0.69) and in the global analysis (0.66 for both protocols). No difference was found in terms of degree of confidence between protocols, for all readers. CONCLUSION An abbreviated MRI protocol for pelvic endometriosis provided an accuracy of interpretation comparable to that of a complete protocol, with similar degrees of confidence and reproducibility, regardless the level of experience.
Collapse
|