Impacto da correção manual sobre a segmentação automatizada do tomógrafo de coerência óptica de domínio espectral

• Ano de defesa: 2017
• Autor(a) principal: Azevedo, Alexandre Gomes Bortoloti de [UNIFESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Paulo (UNIFESP)
• Palavras-chave em Português: Optical coherence tomography; Macula lutea; Optic nerve; Automated segmentation; OCT; Diagnostic techniques, ophthalmological; Nervo óptico; Segmentação automatizada; Tomografia de coerência óptica; Técnicas de diagnóstico oftalmológico
• Link de acesso: https://sucupira.capes.gov.br/sucupira/public/consultas/coleta/trabalhoConclusao/viewTrabalhoConclusao.jsf?popup=true&id_trabalho=5091727; http://repositorio.unifesp.br/handle/11600/50159

OBJECTIVE: To study the automated segmentation of the retinal layers with Spectral domain OCT and the impact of manual correction on segmentation. METHODOLOGY: Manual analysis and correction of automated segmentation of macular thickness and nerve fiber layer of the retrospective optic disc of 551 eyes of 360 consecutive patients using OCT technology Spectralis ™ (Heidelberg Engineering, Heidelberg, Germany). The were examined by two independent examiners and in case of disagreement the examination was reanalyzed together. The nine ETDRS subfields and type of delimitation error of two groups of pathologies macular edema: cystoid macular edema (CME) and neovascular membrane (MNVSR), in addition to patients without macular alterations before and after manual correction. The peripapillary NNRTI exams were also analyzed by two independent examiners and reanalyzed together in case of disagreement. The overall thickness values ​​of the layer of nerve fibers as well as the 8 anatomical segments and their categorizations before and after manual correction. Errors of more than 10% of value were considered of relevant potential impact on clinic RESULTS: There was no significant difference in thickness automated central and manual in normal patients, and with macular edema, while there was a significant difference in central thickness in patients with MNVSR. The frequency of segmentation errors in the macula OCT was In the normal group, 17.14% (24 of 140) in the EMC group and 66.19% (47 of 71) in the MNVSR group. There was a difference greater than 10% initial thickness was 28.16% (20 of 71) in the MNVSR group and 1.4% (two of 140) of the EMC group. There was no error greater than 10% in the normal group. THE frequency of segmentation errors in the nerve fiber layer OCTs in the normal or borderline was 4.16% (seven of 168) and 10.97% (nine of 82) in the category outside the limits of normality (FLN). The frequency of errors in the FLN group was significantly higher (p = 0.038). There was a change of segment in three segments of three patients in the normal and borderline exams and in nine segments of five patients in the group FLN. All exams of the FLN group maintained the same global classification after correction. There was a change of global category only in a normal to borderline. Differences in delimitation responsible for change of only 12 segments in 2000 analyzed segments (0.6%). In the normal and borderline exams, posterior vitreous detachment (DVP) was an important cause of delimitation error (three out of seven cases of error). CONCLUSIONS: After the evaluation of OCT images with segmentation automated procedure, adequate delineation of the macula thickness was patients with normal macular and macular edema and NRWF; lower frequency and magnitude of segmentation errors, with lower impact potential on the interpretation. In the segmentation of macula thickness of patients with MNVSR there was a higher frequency and magnitude of errors, with relevant potential impact on clinical interpretation.