Automatic Detection of Peripheral Retinal Lesions From Ultrawide-Field Fundus Images Using Deep Learning

Peripheral and posterior pole retinal changes in highly myopic Chinese children and adolescents : Retinal changes in Chinese children and adolescents

PURPOSE

This study was conducted to evaluate the prevalence and related factors of peripheral and posterior pole retinal changes in highly myopic Chinese children and adolescents.

METHODS

A hospital-based cross-sectional study was designed. A total of 120 subjects with high myopia were recruited and underwent cycloplegic refraction, dilated fundus examination, and optical coherence tomography. A statistical analysis was performed to evaluate the factors associated with peripheral and posterior pole retinal changes.

RESULTS

The mean spherical equivalent refraction of the subjects was - 8.74 ± 2.86 D, and the mean age was 11.45 ± 3.02 years. Snowflake retinal degeneration (27.5%), white without pressure (27.5%), snail-track degeneration (15%), and lattice degeneration (15%) were the most common peripheral retinal changes, while tessellated fundus (84.17%), optic nerve crescents (78.3%), and posterior staphyloma (11.7%) were the most common posterior changes. Subjects with peripheral changes were significantly older, with thinner choroids (OR = 1.194, 95% CI: 1.045-1.363, p = 0.009; OR = 0.993, 95% CI: 0.987-0.999, p = 0.022, respectively). Optic nerve crescents, tessellated fundus, and posterior scleral staphyloma were all associated with thin choroids (OR = 0.990, 95% CI: 0.983-0.997, p = 0.008; OR = 0.983, 95% CI: 0.974-0.991, p < 0.001; OR = 0.974, 95% CI: 0.960-0.987, p < 0.001, respectively).

CONCLUSION

A substantial proportion of the subjects had peripheral and posterior retinal changes. An increased risk of retinal changes was associated with high degrees of myopia, long axial lengths, thin choroids, and older ages among 7-16-year-old individuals.

Fundus Tessellated Density of Pathologic Myopia

PURPOSE

The study aimed to quantitatively evaluate the fundus tessellated density (FTD) in different categories of pathologic myopia (PM) using fundus photographs with the application of artificial intelligence.

METHODS

A retrospective review of 407 PM (META-PM, Category 2-Category 4) eyes was conducted, employing a biomimetic mechanism of human vision and integrated image processing technologies for FTD extraction and calculation. Different regions of interest were analyzed, including circle O4.5 (optic disc centered, diameter of 4.5 mm) and circle M1.0, M3.0, M6.0 (macular centered, diameter of 1.0, 3.0, and 6.0 mm), using 2 partitioning methods ("X" and "+"). The density of patchy (Category 3) or macular atrophy (Category 4) areas was quantified. Univariate and multivariate linear regression analyses were performed to assess the association with FTD.

RESULTS

The mean FTD of total PM eyes was 0.283, ranging from 0.002 to 0.500, and demonstrating a negative correlation with the PM category. In multivariate analysis, age was found to be significantly associated with FTD ( P <0.05), while axial length did not show a significant association. Fundus tessellation of circle O4.5 and circle M6.0 displayed associations with the FTD across different PM categories. The "X" partitioning method better fit the circle M6.0 region, while both methods were suitable for the circle O4.5 region. After excluding the patchy and macular atrophic areas, the mean FTD values were 0.346 in Category 2, 0.261 in Category 3, and 0.186 in Category 4.

CONCLUSIONS

The study revealed a decreasing trend in FTD values across different categories of PM, regardless of the presence or absence of patchy or macular atrophic areas. Quantifying FTD in PM could be a valuable tool for improving the existing PM classification system and gaining insights into the origin of posterior staphyloma and visual field defects in high myopia.

Automated cervical cell segmentation using deep ensemble learning

BACKGROUND

Cervical cell segmentation is a fundamental step in automated cervical cancer cytology screening. The aim of this study was to develop and evaluate a deep ensemble model for cervical cell segmentation including both cytoplasm and nucleus segmentation.

METHODS

The Cx22 dataset was used to develop the automated cervical cell segmentation algorithm. The U-Net, U-Net +  + , DeepLabV3, DeepLabV3Plus, Transunet, and Segformer were used as candidate model architectures, and each of the first four architectures adopted two different encoders choosing from resnet34, resnet50 and denseNet121. Models were trained under two settings: trained from scratch, encoders initialized from ImageNet pre-trained models and then all layers were fine-tuned. For every segmentation task, four models were chosen as base models, and Unweighted average was adopted as the model ensemble method.

RESULTS

U-Net and U-Net +  + with resnet34 and denseNet121 encoders trained using transfer learning consistently performed better than other models, so they were chosen as base models. The ensemble model obtained the Dice similarity coefficient, sensitivity, specificity of 0.9535 (95% CI:0.9534-0.9536), 0.9621 (0.9619-0.9622),0.9835 (0.9834-0.9836) and 0.7863 (0.7851-0.7876), 0.9581 (0.9573-0.959), 0.9961 (0.9961-0.9962) on cytoplasm segmentation and nucleus segmentation, respectively. The Dice, sensitivity, specificity of baseline models for cytoplasm segmentation and nucleus segmentation were 0.948, 0.954, 0.9823 and 0.750, 0.713, 0.9988, respectively. Except for the specificity of cytoplasm segmentation, all metrics outperformed the best baseline models (P < 0.05) with a moderate margin.

CONCLUSIONS

The proposed algorithm achieved better performances on cervical cell segmentation than baseline models. It can be potentially used in automated cervical cancer cytology screening system.

Spotlight on Lattice Degeneration Imaging Techniques

Lattice degeneration (LD), routinely diagnosed with indirect ophthalmoscopy, is one of the most common and clinically significant peripheral retinal findings. In this review, we have summarized the data on currently available imaging techniques which help to improve diagnosis and our understanding of LD pathogenesis. Ultra-wide field imaging provides reliable color fundus capturing for the primary diagnosis of LD and may also be used as a screening tool. Wide-field imaging can be used for targeted documentation of LD lesions using true colors and with minimal optical distortions. Information on the status of the vitreoretinal interface, including detection of retinal holes, detachments, and vitreous tractions, can be obtained with peripheral structural optical coherence tomography (OCT) or scanning laser ophthalmoscopy in retro-mode. These techniques clarify the associated risks of rhegmatogenous retinal detachment. Fundus autofluorescence can provide details on atrophic changes. However, the risk of retinal detachment by means of this technique requires further investigation. OCT angiography may be successfully performed for some lesions. Taken together, OCT and OCT angiography demonstrate thinning of the choroid, alteration of local choroidal microcirculation, and, in severe lesions, involvement of the sclera. OCT angiography confirms loss of retinal microcirculation within LD lesion, which was previously shown with fluorescein angiography. In conclusion, despite relatively simple primary diagnosis, imaging of LD lesions remains challenging due to their peripheral localization. However, several new strategies, including ultra-wide field imaging, peripheral OCT, and scanning laser ophthalmoscopy, make LD imaging possible on a routine basis, improving diagnosis and understanding of LD pathogenesis.