Perfect cooperative pest control via nano-pesticide and natural predator: High predation selectivity and negligible toxicity toward predatory stinkbug.

The improper use of synthetic pesticides has caused adverse effects on global ecosystems and human health. As a part of sustainable pest management strategy, natural predators, along with nano-pesticides, have made significant contributions to ecological agriculture. The cooperative application of both approaches may overcome their limitations, substantially reducing pesticide application while controlling insect pests efficiently. Herein, the current study introduced a cationic star polymer (SPc) to prepare two types of nano-pesticides, which were co-applied with predatory stinkbugs Picromerus lewisi to achieve perfect cooperative pest control. The SPc exhibited nearly no toxicity against predatory stinkbugs at the working concentration, but it led to the death of predatory stinkbugs at extremely high concentration with the lethal concentration 50 (LC50) value of 13.57 mg/mL through oral feeding method. RNA-seq analysis revealed that the oral feeding of SPc could induce obvious stress responses, leading to stronger phagocytosis, exocytosis, and energy synthesis to ultimately result in the death of predatory stinkbugs. Then, the broflanilide and chlorobenzuron were employed to prepare the self-assembled nano-pesticides via hydrogen bond and Van der Waals force, and the complexation with SPc broke the self-aggregated structures of pesticides and reduced their particle sizes down to nanoscale. The bioactivities of prepared nano-pesticides were significantly improved toward common cutworm Spodoptera litura with the corrected mortality increase by approximately 30%. Importantly, predatory stinkbugs exhibited a strong predation selectivity for alive common cutworms to reduce the exposure risk of nano-pesticides, and the nano-pesticides showed negligible toxicity against predators. Thus, the nano-pesticides and predatory stinkbugs could be applied simultaneously for efficient and sustainable pest management. The current study provides an excellent precedent for perfect cooperative pest control via nano-pesticide and natural predator.

The first A-to-I RNA editome of hemipteran species Coridius chinensis reveals overrepresented recoding and prevalent intron editing in early-diverging insects.

BACKGROUND: Metazoan adenosine-to-inosine (A-to-I) RNA editing resembles A-to-G mutation and increases proteomic diversity in a temporal-spatial manner, allowing organisms adapting to changeable environment. The RNA editomes in many major animal clades remain unexplored, hampering the understanding on the evolution and adaptation of this essential post-transcriptional modification. METHODS: We assembled the chromosome-level genome of Coridius chinensis belonging to Hemiptera, the fifth largest insect order where RNA editing has not been studied yet. We generated ten head RNA-Seq libraries with DNA-Seq from the matched individuals. RESULTS: We identified thousands of high-confidence RNA editing sites in C. chinensis. Overrepresentation of nonsynonymous editing was observed, but conserved recoding across different orders was very rare. Under cold stress, the global editing efficiency was down-regulated and the general transcriptional processes were shut down. Nevertheless, we found an interesting site with "conserved editing but non-conserved recoding" in potassium channel Shab which was significantly up-regulated in cold, serving as a candidate functional site in response to temperature stress. CONCLUSIONS: RNA editing in C. chinensis largely recodes the proteome. The first RNA editome in Hemiptera indicates independent origin of beneficial recoding during insect evolution, which advances our understanding on the evolution, conservation, and adaptation of RNA editing.

Early Predicting Osteogenic Differentiation of Mesenchymal Stem Cells Based on Deep Learning Within One Day.

Osteogenic differentiation of mesenchymal stem cells (MSCs) is proposed to be critical for bone tissue engineering and regenerative medicine. However, the current approach for evaluating osteogenic differentiation mainly involves immunohistochemical staining of specific markers which often can be detected at day 5-7 of osteogenic inducing. Deep learning (DL) is a significant technology for realizing artificial intelligence (AI). Computer vision, a branch of AI, has been proved to achieve high-precision image recognition using convolutional neural networks (CNNs). Our goal was to train CNNs to quantitatively measure the osteogenic differentiation of MSCs. To this end, bright-field images of MSCs during early osteogenic differentiation (day 0, 1, 3, 5, and 7) were captured using a simple optical phase contrast microscope to train CNNs. The results showed that the CNNs could be trained to recognize undifferentiated cells and differentiating cells with an accuracy of 0.961 on the independent test set. In addition, we found that CNNs successfully distinguished differentiated cells at a very early stage (only 1 day). Further analysis showed that overall morphological features of MSCs were the main basis for the CNN classification. In conclusion, MSCs differentiation detection can be achieved early and accurately through simple bright-field images and DL networks, which may also provide a potential and novel method for the field of cell detection in the near future.

Comparative genomic analyses reveal evidence for adaptive A-to-I RNA editing in insect Adar gene.

Although A-to-I RNA editing leads to similar effects to A-to-G DNA mutation, nonsynonymous RNA editing (recoding) is believed to confer its adaptiveness by 'epigenetically' regulating proteomic diversity in a temporospatial manner, avoiding the pleiotropic effect of genomic mutations. Recent discoveries on the evolutionary trajectory of Ser>Gly auto-editing site in insect Adar gene demonstrated a selective advantage to having an editable codon compared to uneditable ones. However, apart from pure observations, quantitative approaches for justifying the adaptiveness of individual RNA editing sites are still lacking. We performed a comparative genomic analysis on 113 Diptera species, focusing on the Adar Ser>Gly auto-recoding site in Drosophila. We only found one species having a derived Gly at the corresponding site, and this occurrence was significantly lower than genome-wide random expectation. This suggests that the Adar Ser>Gly site is unlikely to be genomically replaced with G during evolution, and thus indicating the advantage of editable status over hardwired genomic alleles. Similar trends were observed for the conserved Ile>Met recoding in gene Syt1. In the light of evolution, we established a comparative genomic approach for quantitatively justifying the adaptiveness of individual editing sites. Priority should be given to such adaptive editing sites in future functional studies.

Osmotic Pressure and Its Biological Implications.

Gaining insight into osmotic pressure and its biological implications is pivotal for revealing mechanisms underlying numerous fundamental biological processes across scales and will contribute to the biomedical and pharmaceutical fields. This review aims to provide an overview of the current understanding, focusing on two central issues: (i) how to determine theoretically osmotic pressure and (ii) how osmotic pressure affects important biological activities. More specifically, we discuss the representative theoretical equations and models for different solutions, emphasizing their applicability and limitations, and summarize the effect of osmotic pressure on lipid phase separation, cell division, and differentiation, focusing on the mechanisms underlying the osmotic pressure dependence of these biological processes. We highlight that new theory of osmotic pressure applicable for all experimentally feasible temperatures and solute concentrations needs to be developed, and further studies regarding the role of osmotic pressure in other biological processes should also be carried out to improve our comprehensive and in-depth understanding. Moreover, we point out the importance and challenges of developing techniques for the in vivo measurement of osmotic pressure.

A new ant species of the genus Carebara Westwood, 1840 (Hymenoptera, Formicidae, Myrmicinae) with a key to Chinese species.

A new Chinese ant species Carebaralaevicepssp. nov. is described based on the major and minor workers. This species is most similar to C.lusciosa (Wheeler, 1928) due to a spineless propodeum, the absence of horns, and a smooth head capsule. It is distinguished by the following features: (1) antenna 10-segmented; (2) katepisternum rugose-reticulate; (3) in major workers, lateral sides of head in full-face view parallel; (4) metanotal groove distinct, anterodorsal corner forming an acute tooth behind metanotal groove. Moreover, an updated key to Chinese Carebara species is presented based on major workers, with a checklist comprising a total of 36 Chinese Carebara species and subspecies. Morphological structures and scanning electron micrographs of the newly discovered species' minor and major workers are provided.

Translating color fundus photography to indocyanine green angiography using deep-learning for age-related macular degeneration screening.

Age-related macular degeneration (AMD) is the leading cause of central vision impairment among the elderly. Effective and accurate AMD screening tools are urgently needed. Indocyanine green angiography (ICGA) is a well-established technique for detecting chorioretinal diseases, but its invasive nature and potential risks impede its routine clinical application. Here, we innovatively developed a deep-learning model capable of generating realistic ICGA images from color fundus photography (CF) using generative adversarial networks (GANs) and evaluated its performance in AMD classification. The model was developed with 99,002 CF-ICGA pairs from a tertiary center. The quality of the generated ICGA images underwent objective evaluation using mean absolute error (MAE), peak signal-to-noise ratio (PSNR), structural similarity measures (SSIM), etc., and subjective evaluation by two experienced ophthalmologists. The model generated realistic early, mid and late-phase ICGA images, with SSIM spanned from 0.57 to 0.65. The subjective quality scores ranged from 1.46 to 2.74 on the five-point scale (1 refers to the real ICGA image quality, Kappa 0.79-0.84). Moreover, we assessed the application of translated ICGA images in AMD screening on an external dataset (n = 13887) by calculating area under the ROC curve (AUC) in classifying AMD. Combining generated ICGA with real CF images improved the accuracy of AMD classification with AUC increased from 0.93 to 0.97 (P < 0.001). These results suggested that CF-to-ICGA translation can serve as a cross-modal data augmentation method to address the data hunger often encountered in deep-learning research, and as a promising add-on for population-based AMD screening. Real-world validation is warranted before clinical usage.

The complete mitochondrial genome of Aphidius colemani (Hymenoptera: Braconidae: Aphidiinae).

The genome-level features are crucial genetic resources for species identification and phylogenetic analysis. Here, the complete mitochondrial genome of Aphidius colemani Viereck 1912 (Hymenoptera: Braconidae: Aphidiinae) was sequenced, determined and analyzed. The circular genome is 16,372 bp in length with an overall base composition of 38.9% for A, 46.2% for T, 6.7% for C, and 8.2% for G. The mitochondrial genome of A. colemani contained 13 protein-coding genes that initiated by the ATN codon, 22 transfer RNA genes, two ribosomal RNA genes (rRNAs), and a control region (CR). It shared the same gene arrangement patterns that occurred in two tRNA clusters of trnI-trnQ-trnM and trnW-trnC-trnY with Aphidius gifuensis. Phylogenetic analyses using Bayesian inference and Maximum-likelihood methods supported that the two species of Aphidiinae formed a clade and sister to other subfamilies of Braconidae.

Chromosome-level genome of the poultry shaft louse Menopon gallinae provides insight into the host-switching and adaptive evolution of parasitic lice.

BACKGROUND: Lice (Psocodea: Phthiraptera) are one important group of parasites that infects birds and mammals. It is believed that the ancestor of parasitic lice originated on the ancient avian host, and ancient mammals acquired these parasites via host-switching from birds. Here we present the first chromosome-level genome of Menopon gallinae in Amblycera (earliest diverging lineage of parasitic lice). We explore the transition of louse host-switching from birds to mammals at the genomic level by identifying numerous idiosyncratic genomic variations. RESULTS: The assembled genome is 155 Mb in length, with a contig N50 of 27.42 Mb. Hi-C scaffolding assigned 97% of the bases to 5 chromosomes. The genome of M. gallinae retains a basal insect repertoire of 11,950 protein-coding genes. By comparing the genomes of lice to those of multiple representative insects in other orders, we discovered that gene families of digestion, detoxification, and immunity-related are generally conserved between bird lice and mammal lice, while mammal lice have undergone a significant reduction in genes related to chemosensory systems and temperature. This suggests that mammal lice have lost some of these genes through the adaption to environment and temperatures after host-switching. Furthermore, 7 genes related to hematophagy were positively selected in mammal lice, suggesting their involvement in the hematophagous behavior. CONCLUSIONS: Our high-quality genome of M. gallinae provides a valuable resource for comparative genomic research in Phthiraptera and facilitates further studies on adaptive evolution of host-switching within parasitic lice.

Stem chewing lice on Cretaceous feathers preserved in amber.

Phthirapteran lice (true lice or parasitic lice) are a major group of ectoparasitic insects living on their bird or mammal hosts during their entire life cycle.1 Due to their highly specialized lifestyles, they are extremely poorly represented in fossil records.2 Molecular clock estimations have speculated extensively about the origin time of parasitic lice,3,4 yet none have been confirmed unequivocally. Herein, we report a new family of stem chewing lice, based on two adult insects associated with several semiplume feathers preserved within a piece of Kachin amber from the mid-Cretaceous. They display some defining characteristics of the Amblycera, an early-diverging lineage of the crown lice group. These features include a wingless body, chewing mouthparts, narrow and small thorax, and short tarsus with elongated euplantulae. Our phylogenetic analysis places the new taxa in the Amblycera, and the discovery thus pushes back the lice fossil records by at least 55 million years. Furthermore, the new specimens show primitive characters such as compressed and club-shaped terminal segments of antennae, maxillary and labial palps, and unmodified femora of hind legs, providing key information for the evolutionary relationship between free-living booklice and parasitic lice. This suggests that some ectoparasitic characters defining the crown lice group might have evolved among amblyceran and non-amblyceran lice in parallel. These newly described fossil specimens imply at least a Cretaceous age of Phthiraptera.

Natural selection and genetic diversity maintenance in a parasitic wasp during continuous biological control application.

Aphidius gifuensis is a parasitoid wasp and primary endoparasitoid enemy of the peach potato aphid, Myzus persicae. Artificially reared, captive wasps of this species have been extensively and effectively used to control populations of aphids and limit crop loss. However, the consequences of large-scale releasing of captive A. gifuensis, such as genetic erosion and reduced fitness in wild populations of this species, remains unclear. Here, we sequence the genomes of 542 A. gifuensis individuals collected across China, including 265 wild and 277 human-intervened samples. Population genetic analyses on wild individuals recovered Yunnan populations as the ancestral group with the most complex genetic structure. We also find genetic signature of environmental adaptation during the dispersal of wild populations from Yunnan to other regions. While comparative genomic analyses of captive wasps revealed a decrease in genetic diversity during long-term rearing, population genomic analyses revealed signatures of natural selection by several biotic (host plants) or abiotic (climate) factors, which support maintenance of the gene pool of wild populations in spite of the introduction of captive wasps. Therefore, the impact of large-scale release is reduced. Our study suggests that A. gifuensis is a good system for exploring the genetic and evolutionary effects of mass rearing and release on species commonly used as biocontrol agents.

Matrix metalloproteinases as attractive therapeutic targets for chronic pain: A narrative review.

Chronic pain constitutes an abnormal pain state that detrimentally affects the quality of life, daily activities, occupational performance, and stability of mood. Despite the prevalence of chronic pain, effective drugs with potent abirritation and minimal side effects remain elusive. Substantial studies have revealed aberrant activation of the matrix metalloproteinases (MMPs) in multiple chronic pain models. Additionally, emerging evidence has demonstrated that the downregulation of MMPs can alleviate chronic pain in diverse animal models, underscoring the unique and crucial role of MMPs in different stages and types of chronic pain. This review delves into the mechanistic insights and roles of MMPs in modulating chronic pain. The aberrant activation of MMPs has been linked to neuropathic pain through mechanisms involving myelin abnormalities in peripheral nerve and spinal dorsal horn (SDH), hyperexcitability of dorsal root ganglion (DRG) neurons, activation of N-methyl-d-aspartate receptors (NMDAR) and Ca2+-dependent signals, glial cell activation, and proinflammatory cytokines release. Different MMPs also contribute significantly to inflammatory pain and cancer pain. Furthermore, we summarized the substantial therapeutic potential of MMP pharmacological inhibitors across different types of chronic pain. Overall, our findings underscore the promising therapeutic prospects of MMPs targeting for managing chronic pain.

Interleukin-17: A Putative Novel Pharmacological Target for Pathological Pain.

Pathological pain imposes a huge burden on the economy and the lives of patients. At present, drugs used for the treatment of pathological pain have only modest efficacy and are also plagued by adverse effects and risk for misuse and abuse. Therefore, understanding the mechanisms of pathological pain is essential for the development of novel analgesics. Several lines of evidence indicate that interleukin-17 (IL-17) is upregulated in rodent models of pathological pain in the periphery and central nervous system. Besides, the administration of IL-17 antibody alleviated pathological pain. Moreover, IL-17 administration led to mechanical allodynia which was alleviated by the IL-17 antibody. In this review, we summarized and discussed the therapeutic potential of targeting IL-17 for pathological pain. The upregulation of IL-17 promoted the development of pathological pain by promoting neuroinflammation, enhancing the excitability of dorsal root ganglion neurons, and promoting the communication of glial cells and neurons in the spinal cord. In general, the existing research shows that IL-17 is an attractive therapeutic target for pathologic pain, but the underlying mechanisms still need to be investigated.

Targeting the JAK2/STAT3 signaling pathway for chronic pain.

Chronic pain is a notable health concern because of its prevalence, persistence, and associated mental stress. Drugs targeting chronic pain with potent abirritation, and minimal side effects remain unidentified. Substantial evidence indicates that the Janus Kinase 2 (JAK2)/signal transducer and activator of transcription 3 (STAT3) signaling pathway plays a distinct and critical role in different stages of chronic pain. Aberrant activation of the JAK2/STAT3 signaling pathway is evident in multiple chronic pain models. Moreover, an increasing number of studies have demonstrated that the downregulation of JAK2/STAT3 can attenuate chronic pain in different animal models. In this review, we investigated the mechanism and role of the JAK2/STAT3 signaling pathway in modulating chronic pain. The aberrant activation of JAK2/STAT3 can trigger chronic pain by interacting with microglia and astrocytes, releasing proinflammatory cytokines, inhibiting anti-inflammatory cytokines, and regulating synaptic plasticity. We also retrospectively reviewed current reports on JAK2/STAT3 pharmacological inhibitors that demonstrated their significant therapeutic potential in different types of chronic pain. In summary, our results provide strong evidence that the JAK2/STAT3 signaling pathway is a promising therapeutic target for chronic pain.

SI-ViT: Shuffle instance-based Vision Transformer for pancreatic cancer ROSE image classification.

BACKGROUND AND OBJECTIVE: The rapid on-site evaluation (ROSE) technique improves pancreatic cancer diagnosis by enabling immediate analysis of fast-stained cytopathological images. Automating ROSE classification could not only reduce the burden on pathologists but also broaden the application of this increasingly popular technique. However, this approach faces substantial challenges due to complex perturbations in color distribution, brightness, and contrast, which are influenced by various staining environments and devices. Additionally, the pronounced variability in cancerous patterns across samples further complicates classification, underscoring the difficulty in precisely identifying local cells and establishing their global relationships. METHODS: To address these challenges, we propose an instance-aware approach that enhances the Vision Transformer with a novel shuffle instance strategy (SI-ViT). Our approach presents a shuffle step to generate bags of shuffled instances and corresponding bag-level soft-labels, allowing the model to understand relationships and distributions beyond the limited original distributions. Simultaneously, combined with an un-shuffle step, the traditional ViT can model the relationships corresponding to the sample labels. This dual-step approach helps the model to focus on inner-sample and cross-sample instance relationships, making it potent in extracting diverse image patterns and reducing complicated perturbations. RESULTS: Compared to state-of-the-art methods, significant improvements in ROSE classification have been achieved. Aiming for interpretability, equipped with instance shuffling, SI-ViT yields precise attention regions that identifying cancer and normal cells in various scenarios. Additionally, the approach shows excellent potential in pathological image analysis through generalization validation on other datasets. CONCLUSIONS: By proposing instance relationship modeling through shuffling, we introduce a new insight in pathological image analysis. The significant improvements in ROSE classification leads to protential AI-on-site applications in pancreatic cancer diagnosis. The code and results are publicly available at https://github.com/sagizty/MIL-SI.

A deep learning model for generating fundus autofluorescence images from color fundus photography.

Background: Fundus Autofluorescence (FAF) is a valuable imaging technique used to assess metabolic alterations in the retinal pigment epithelium (RPE) associated with various age-related and disease-related changes. The practical uses of FAF are ever-growing. This study aimed to evaluate the effectiveness of a generative deep learning (DL) model in translating color fundus (CF) images into synthetic FAF images and explore its potential for enhancing screening of age-related macular degeneration (AMD). Methods: A generative adversarial network (GAN) model was trained on pairs of CF and FAF images to generate synthetic FAF images. The quality of synthesized FAF images was assessed objectively by common generation metrics. Additionally, the clinical effectiveness of the generated FAF images in AMD classification was evaluated by measuring the area under the curve (AUC), using the LabelMe dataset. Results: A total of 8410 FAF images from 2586 patients were analyzed. The synthesized FAF images exhibited an impressive objectively assessed quality, achieving a multi-scale structural similarity index (MS-SSIM) of 0.67. When evaluated on the LabelMe dataset, the combination of generated FAF images and CF images resulted in a noteworthy improvement in AMD classification accuracy, with the AUC increasing from 0.931 to 0.968. Conclusions: This study presents the first attempt to use a generative deep learning model to create authentic and high-quality FAF images from CF images. The incorporation of the translated FAF images on top of CF images improved the accuracy of AMD classification. Overall, this study presents a promising approach to enhance large-scale AMD screening.

Genetic Modification of a Hox Locus Drives Mimetic Color Pattern Variation in a Highly Polymorphic Bumble Bee.

Müllerian mimicry provides natural replicates ideal for exploring mechanisms underlying adaptive phenotypic divergence and convergence, yet the genetic mechanisms underlying mimetic variation remain largely unknown. The current study investigates the genetic basis of mimetic color pattern variation in a highly polymorphic bumble bee, Bombus breviceps (Hymenoptera, Apidae). In South Asia, this species and multiple comimetic species converge onto local Müllerian mimicry patterns by shifting the abdominal setal color from orange to black. Genetic crossing between the orange and black phenotypes suggested the color dimorphism being controlled by a single Mendelian locus, with the orange allele being dominant over black. Genome-wide association suggests that a locus at the intergenic region between 2 abdominal fate-determining Hox genes, abd-A and Abd-B, is associated with the color change. This locus is therefore in the same intergenic region but not the same exact locus as found to drive red black midabdominal variation in a distantly related bumble bee species, Bombus melanopygus. Gene expression analysis and RNA interferences suggest that differential expression of an intergenic long noncoding RNA between abd-A and Abd-B at the onset setal color differentiation may drive the orange black color variation by causing a homeotic shift late in development. Analysis of this same color locus in comimetic species reveals no sequence association with the same color shift, suggesting that mimetic convergence is achieved through distinct genetic routes. Our study establishes Hox regions as genomic hotspots for color pattern evolution in bumble bees and demonstrates how pleiotropic developmental loci can drive adaptive radiations in nature.

Identification and Interpretation of A-to-I RNA Editing Events in Insect Transcriptomes.

Adenosine-to-inosine (A-to-I) RNA editing is the most prevalent RNA modification in the nervous systems of metazoans. To study the biological significance of RNA editing, we first have to accurately identify these editing events from the transcriptome. The genome-wide identification of RNA editing sites remains a challenging task. In this review, we will first introduce the occurrence, regulation, and importance of A-to-I RNA editing and then describe the established bioinformatic procedures and difficulties in the accurate identification of these sit esespecially in small sized non-model insects. In brief, (1) to obtain an accurate profile of RNA editing sites, a transcriptome coupled with the DNA resequencing of a matched sample is favorable; (2) the single-cell sequencing technique is ready to be applied to RNA editing studies, but there are a few limitations to overcome; (3) during mapping and variant calling steps, various issues, like mapping and base quality, soft-clipping, and the positions of mismatches on reads, should be carefully considered; (4) Sanger sequencing of both RNA and the matched DNA is the best verification of RNA editing sites, but other auxiliary evidence, like the nonsynonymous-to-synonymous ratio or the linkage information, is also helpful for judging the reliability of editing sites. We have systematically reviewed the understanding of the biological significance of RNA editing and summarized the methodology for identifying such editing events. We also raised several promising aspects and challenges in this field. With insightful perspectives on both scientific and technical issues, our review will benefit the researchers in the broader RNA editing community.

FTO Sensitizes Oral Squamous Cell Carcinoma to Ferroptosis via Suppressing ACSL3 and GPX4.

Ferroptosis is a newly established form of regulated cell death characterized by intracellular lipid peroxidation and iron accumulation that may be a promising cancer treatment strategy. However, the function and therapeutic value of ferroptosis in oral squamous cell carcinoma (OSCC) remain inadequately understood. In the present study, we investigated the biological role of the fat mass and obesity-associated gene (FTO) in ferroptosis in the context of OSCC. We found that OSCC had greater potential for ferroptosis, and FTO is associated with ferroptosis. Furthermore, higher FTO expression sensitized OSCC cells to ferroptosis in vitro and in vivo. Mechanistically, FTO suppressed the expression of anti-ferroptotic factors, acyl-CoA synthetase long-chain family member 3 (ACSL3) and glutathione peroxidase 4 (GPX4), by demethylating the m6A modification on the mRNA of ACSL3 and GPX4 and decreasing their stability. Taken together, our findings revealed that FTO promotes ferroptosis through ACSL3 and GPX4 regulation. Thus, ferroptosis activation in OSCC with high FTO levels may serve as a potential therapeutic target.

Translation of Color Fundus Photography into Fluorescein Angiography Using Deep Learning for Enhanced Diabetic Retinopathy Screening.

Purpose: To develop and validate a deep learning model that can transform color fundus (CF) photography into corresponding venous and late-phase fundus fluorescein angiography (FFA) images. Design: Cross-sectional study. Participants: We included 51 370 CF-venous FFA pairs and 14 644 CF-late FFA pairs from 4438 patients for model development. External testing involved 50 eyes with CF-FFA pairs and 2 public datasets for diabetic retinopathy (DR) classification, with 86 952 CF from EyePACs, and 1744 CF from MESSIDOR2. Methods: We trained a deep-learning model to transform CF into corresponding venous and late-phase FFA images. The translated FFA images' quality was evaluated quantitatively on the internal test set and subjectively on 100 eyes with CF-FFA paired images (50 from external), based on the realisticity of the global image, anatomical landmarks (macula, optic disc, and vessels), and lesions. Moreover, we validated the clinical utility of the translated FFA for classifying 5-class DR and diabetic macular edema (DME) in the EyePACs and MESSIDOR2 datasets. Main Outcome Measures: Image generation was quantitatively assessed by structural similarity measures (SSIM), and subjectively by 2 clinical experts on a 5-point scale (1 refers real FFA); intragrader agreement was assessed by kappa. The DR classification accuracy was assessed by area under the receiver operating characteristic curve. Results: The SSIM of the translated FFA images were > 0.6, and the subjective quality scores ranged from 1.37 to 2.60. Both experts reported similar quality scores with substantial agreement (all kappas > 0.8). Adding the generated FFA on top of CF improved DR classification in the EyePACs and MESSIDOR2 datasets, with the area under the receiver operating characteristic curve increased from 0.912 to 0.939 on the EyePACs dataset and from 0.952 to 0.972 on the MESSIDOR2 dataset. The DME area under the receiver operating characteristic curve also increased from 0.927 to 0.974 in the MESSIDOR2 dataset. Conclusions: Our CF-to-FFA framework produced realistic FFA images. Moreover, adding the translated FFA images on top of CF improved the accuracy of DR screening. These results suggest that CF-to-FFA translation could be used as a surrogate method when FFA examination is not feasible and as a simple add-on to improve DR screening. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.