Objective To compare the differences between swept-source optical coherence tomography angiography(SS-OCTA)and ultra-wide-field fluorescein angiography(UWFA)in detecting non-perfusion areas(NPs)in patients with diabetic retinopathy(DR),to evaluate the accuracy of SS-OCTA in predicting NPs outside its visible range,and to explore the distribution patterns of NPs.Methods A retrospective analysis was made on 69 DR patients(88 eyes)who under-went both UWFA and SS-OCTA examinations at the Ophthalmology Department of Sichuan Provincial People's Hospital from December 2022 to September 2024.Manual NP labeling was conducted to compare the detection rate of NPs between the two imaging techniques.The distribution patterns of NPs and the accuracy of SS-OCTA for predicting NPs outside its visible range were also analyzed.Results In a scanning area of 20 mm x 24 mm,the overall NP detection rate by SS-OCTA was 47.40％,with UWFA taken as the standard.The NP detection rate by SS-OCTA was 51.56％in the superotemporal quad-rant,58.35％in the inferotemporal quadrant,45.50％in the superonasal quadrant,and 43.17％in the inferonasal quad-rant.Most NPs occurred in the inferonasal quadrant,accounting for 41.71％of the total NP.The accuracy of SS-OCTA in predicting NPs was 75.00％in the superonasal quadrant and 78.41％in the inferonasal quadrant.The ischemic indices(ISI)of the two imaging techniques were highly positively correlated(r2=0.74).Conclusion Although SS-OCTA can-not yet fully replace UWFA for NP detection in DR patients due to a small visible range,it is still an effective tool to assess retinal ischemia.SS-OCTA has the ability to predict NPs outside its visible range in its scanning range.The inferonasal quadrant is the region where NPs occur most frequently in DR patients,so it is suggested that special attention should be paid to this region in early diagnosis and follow-up periods.

Objective Deep learning provides strong technical support for early diagnosis,lesion segmentation,and treatment prediction of retinal diseases,significantly improving the efficiency and accuracy of diagnosis.But it also faces challenges in terms of different applicability and performance differences of the model,mainly due to the differences in fea-ture extraction ability,computational complexity,and clinical adaptability among different network structures,which make them have different advantages and limitations in different application scenarios.By systematically searching relevant litera-ture in PubMed and Web of Science databases over the past 5 years,this article summarizes the most commonly used deep learning network architectures in common vitreoretinal diseases,summarizes their different advantages and limitations,and analyzes the best application directions of each architecture in the field of ophthalmology,providing reference and inspira-tion for future research.

Objective To compare the differences between swept-source optical coherence tomography angiography(SS-OCTA)and ultra-wide-field fluorescein angiography(UWFA)in detecting non-perfusion areas(NPs)in patients with diabetic retinopathy(DR),to evaluate the accuracy of SS-OCTA in predicting NPs outside its visible range,and to explore the distribution patterns of NPs.Methods A retrospective analysis was made on 69 DR patients(88 eyes)who under-went both UWFA and SS-OCTA examinations at the Ophthalmology Department of Sichuan Provincial People's Hospital from December 2022 to September 2024.Manual NP labeling was conducted to compare the detection rate of NPs between the two imaging techniques.The distribution patterns of NPs and the accuracy of SS-OCTA for predicting NPs outside its visible range were also analyzed.Results In a scanning area of 20 mm x 24 mm,the overall NP detection rate by SS-OCTA was 47.40％,with UWFA taken as the standard.The NP detection rate by SS-OCTA was 51.56％in the superotemporal quad-rant,58.35％in the inferotemporal quadrant,45.50％in the superonasal quadrant,and 43.17％in the inferonasal quad-rant.Most NPs occurred in the inferonasal quadrant,accounting for 41.71％of the total NP.The accuracy of SS-OCTA in predicting NPs was 75.00％in the superonasal quadrant and 78.41％in the inferonasal quadrant.The ischemic indices(ISI)of the two imaging techniques were highly positively correlated(r2=0.74).Conclusion Although SS-OCTA can-not yet fully replace UWFA for NP detection in DR patients due to a small visible range,it is still an effective tool to assess retinal ischemia.SS-OCTA has the ability to predict NPs outside its visible range in its scanning range.The inferonasal quadrant is the region where NPs occur most frequently in DR patients,so it is suggested that special attention should be paid to this region in early diagnosis and follow-up periods.

Objective Deep learning provides strong technical support for early diagnosis,lesion segmentation,and treatment prediction of retinal diseases,significantly improving the efficiency and accuracy of diagnosis.But it also faces challenges in terms of different applicability and performance differences of the model,mainly due to the differences in fea-ture extraction ability,computational complexity,and clinical adaptability among different network structures,which make them have different advantages and limitations in different application scenarios.By systematically searching relevant litera-ture in PubMed and Web of Science databases over the past 5 years,this article summarizes the most commonly used deep learning network architectures in common vitreoretinal diseases,summarizes their different advantages and limitations,and analyzes the best application directions of each architecture in the field of ophthalmology,providing reference and inspira-tion for future research.