Search

Abstract This study introduces an innovative framework named scale for processing dynamic skyline queries securely in cloud environments. Unlike previous approaches that require complex operations on encrypted data, scale simplifies dynamic skyline domination to mere comparisons, significantly improving query efficiency. Through empirical evaluations over four datasets, we show that scale accelerates query processing nearly 1000-fold compared to existing state-of-the-art methods. Specifically, scale shows significant efficiency improvements by simplifying query interactions to a single round between the user and the cloud, which is validated through empirical studies on multiple datasets. Moreover, we introduce two distributed versions of scale, dist-scale-s and dist-scale-e, which further optimize performance by facilitating parallel processing. This adaptation showcases a substantial reduction in response times and computational overhead, underpinning the scalability and effectiveness of our framework in handling large-scale, secure cloud-based queries.

Abstract Background Resistance to ferroptosis, a regulated cell death caused by iron‐dependent excessive accumulation of lipid peroxides, has recently been linked to lung adenocarcinoma (LUAD). Intracellular antioxidant systems are required for protection against ferroptosis. The purpose of the present study was to investigate whether and how extracellular system desensitizes LUAD cells to ferroptosis. Methods Established human lung fibroblasts MRC‐5, WI38, and human LUAD H1650, PC9, H1975, H358, A549, and H1299 cell lines, tumor and matched normal adjacent tissues of LUAD, and plasma from healthy individuals and LUAD patients were used in this study. Immunohistochemistry and immunoblotting were used to analyze protein expression, and quantitative reverse transcription‐PCR was used to analyze mRNA expression. Cell viability, cell death, and the lipid reactive oxygen species generation were measured to evaluate the responses to ferroptosis. Exosomes were observed using transmission electron microscope. The localization of arachidonic acid (AA) was detected using click chemistry labeling followed by confocal microscopy. Interactions between RNAs and proteins were detected using RNA pull‐down, RNA immunoprecipitation and photoactivatable ribonucleoside‐enhanced crosslinking and immunoprecipitation methods. Proteomic analysis was used to investigate RNA‐regulated proteins, and metabolomic analysis was performed to analyze metabolites. Cell‐derived xenograft, patient‐derived xenograft, cell‐implanted intrapulmonary LUAD mouse models and plasma/tissue specimens from LUAD patients were used to validate the molecular mechanism. Results Plasma exosome from LUAD patients specifically reduced lipid peroxidation and desensitized LUAD cells to ferroptosis. A potential explanation is that exosomal circRNA_101093 (cir93) maintained an elevation in intracellular cir93 in LUAD to modulate AA, a poly‐unsaturated fatty acid critical for ferroptosis‐associated increased peroxidation in the plasma membrane. Mechanistically, cir93 interacted with and increased fatty acid‐binding protein 3 (FABP3), which transported AA and facilitated its reaction with taurine. Thus, global AA was reduced, whereas N‐arachidonoyl taurine (NAT, the product of AA and taurine) was induced. Notably, the role of NAT in suppressing AA incorporation into the plasma membrane was also revealed. In pre‐clinical in vivo models, reducing exosome improved ferroptosis‐based treatment. Conclusion Exosome and cir93 are essential for desensitizing LUAD cells to ferroptosis, and blocking exosome may be helpful for future LUAD treatment.

Abstract Yes-associated protein (YAP) activation is crucial for tumor formation and development, and its stability is regulated by ubiquitination. ISGylation is a type of ubiquitination like post-translational modification, whereas whether YAP is ISGylated and how ISGylation influences YAP ubiquitination-related function remains uncovered. In addition, YAP can activate glucose metabolism by activating the hexosamine biosynthesis pathway (HBP) and glycolysis, and generate a large number of intermediates to promote tumor proliferation. However, whether YAP stimulates the pentose phosphate pathway (PPP), another tumor-promoting glucose metabolism pathway, and the relationship between this stimulation and ISGylation needs further investigation. Here, we found that YAP was ISGylated and this ISGylation inhibited YAP ubiquitination, proteasome degradation, interaction with-beta-transducin repeat containing E3 ubiquitin-protein ligase (βTrCP) to promote YAP stability. However, ISGylation-induced pro-YAP effects were abolished by YAP K497R (K, lysine; R, arginine) mutation, suggesting K497 could be the major YAP ISGylation site. In addition, YAP ISGylation promoted cell viability, cell-derived xenograft (CDX) and patient-derived xenograft (PDX) tumor formation. YAP ISGylation also increased downstream genes transcription, including one of the key enzymes of PPP, 6-phosphogluconolactonase (6PGL). Mechanistically, YAP promoted 6PGL transcription by simultaneously recruiting SMAD family member 2 (SMAD2) and TEA domain transcription factor 4 (TEAD4) binding to the 6PGL promoter to activate PPP. In clinical lung adenocarcinoma (LUAD) specimens, we found that YAP ISGylation degree was positively associated with 6PGL mRNA level, especially in high glucose LUAD tissues compared to low glucose LUAD tissues. Collectively, this study suggested that YAP ISGylation is critical for maintaining its stability and further activation of PPP. Targeting ISGylated YAP might be a new choice for hyperglycemia cancer treatment.

Abstract Background Lung adenocarcinoma (LUAD) is the most common subtype of lung cancer. Patient prognosis is poor, and the existing therapeutic strategies for LUAD are far from satisfactory. Recently, targeting N6-methyladenosine (m6A) modification of RNA has been suggested as a potential strategy to impede tumor progression. However, the roles of m6A modification in LUAD tumorigenesis is unknown. Methods Global m6A levels and expressions of m6A writers, erasers and readers were evaluated by RNA methylation assay, dot blot, immunoblotting, immunohistochemistry and ELISA in human LUAD, mouse models and cell lines. Cell viability, 3D-spheroid generation, in vivo LUAD formation, experiments in cell- and patient-derived xenograft mice and survival analysis were conducted to explore the impact of m6A on LUAD. The RNA-protein interactions, translation, putative m6A sites and glycolysis were explored in the investigation of the mechanism underlying how m6A stimulates tumorigenesis. Results The elevation of global m6A level in most human LUAD specimens resulted from the combined upregulation of m6A writer methyltransferase 3 (METTL3) and downregulation of eraser alkB homolog 5 (ALKBH5). Elevated global m6A level was associated with a poor overall survival in LUAD patients. Reducing m6A levels by knocking out METTL3 and overexpressing ALKBH5 suppressed 3D-spheroid generation in LUAD cells and intra-pulmonary tumor formation in mice. Mechanistically, m6A-dependent stimulation of glycolysis and tumorigenesis occurred via enolase 1 (ENO1). ENO1 mRNA was m6A methylated at 359 A, which facilitated it’s binding with the m6A reader YTH N6-methyladenosine RNA binding protein 1 (YTHDF1) and resulted in enhanced translation of ENO1. ENO1 positively correlated with METTL3 and global m6A levels, and negatively correlated with ALKBH5 in human LUAD. In addition, m6A-dependent elevation of ENO1 was associated with LUAD progression. In preclinical models, tumors with a higher global m6A level showed a more sensitive response to the inhibition of pan-methylation, glycolysis and ENO activity in LUAD. Conclusions The m6A-dependent stimulation of glycolysis and tumorigenesis in LUAD is at least partially orchestrated by the upregulation of METTL3, downregulation of ALKBH5, and stimulation of YTHDF1-mediated ENO1 translation. Blocking this mechanism may represent a potential treatment strategy for m6A-dependent LUAD.

A lack of promising targets leads to poor prognosis in patients with lung adenocarcinoma (LUAD). Therefore, it is urgent to identify novel therapeutic targets. The importance of the N6-methyladenosine (m6A) RNA modification has been demonstrated in various types of tumors; however, knowledge of m6A-related proteins in LUAD is still limited. Here, we found that insulin-like growth factor 2 mRNA binding protein 3 (IGF2BP3), an m6A reader protein, is highly expressed in LUAD and associated with poor prognosis. IGF2BP3 desensitizes ferroptosis (a new form of regulated cell death) in a manner dependent on its m6A reading domain and binding capacity to m6A-methylated mRNAs encoding anti-ferroptotic factors, including but not limited to glutathione peroxidase 4 (GPX4), solute carrier family 3 member 2 (SLC3A2), acyl-CoA synthetase long chain family member 3 (ACSL3), and ferritin heavy chain 1 (FTH1). After IGF2BP3 overexpression, expression levels and mRNA stabilities of these anti-ferroptotic factors were successfully sustained. Notably, significant correlations between SLC3A2, ACSL3, and IGF2BP3 were revealed in clinical LUAD specimens, further establishing the essential role of IGF2BP3 in desensitizing ferroptosis. Inducing ferroptosis has been gradually accepted as an alternative strategy to treat tumors. Thus, IGF2BP3 could be a potential target for the future development of new biomaterial-associated therapeutic anti-tumor drugs.

Abstract Tribbles homolog 2 (TRIB2) is known to boost liver tumorigenesis via regulating Ubiquitin (Ub) proteasome system (UPS). At least two ways are involved, i.e., acts as an adaptor protein to modulate ubiquitination functions of certain ubiquitin E3 ligases (E3s) and reduces global Ub levels via increasing the proteolysis activity of proteasome. Recently, we have identified the role of TRIB2 to relieve oxidative damage via reducing the availability of Ub that is essential for the ubiquitination and subsequent degradation of Glutathione peroxidase 4 (GPX4). Although GPX4 is a critical antioxidant factor to protect against ferroptosis, the exact evidence showing that TRIB2 desensitizes ferroptosis is lacking. Also, whether such function is via E3 remains unclear. Here, we demonstrated that deletion of TRIB2 sensitized ferroptosis via lifting labile iron in liver cancer cells. By contrast, overexpression of TRIB2 led to the opposite outcome. We further demonstrated that transferrin receptor (TFRC) was required for TRIB2 to desensitize the cells to ferroptosis. Without TFRC, the labile iron pool could not be reduced by overexpressing TRIB2. We also found that beta-transducin repeat containing E3 ubiqutin protein ligase (βTrCP) was a genuine E3 for the ubiquitination of TFRC, and TRIB2 was unable to decline labile iron level once upon βTrCP was knocked out. In addition, we confirmed that the opposite effects on ferroptosis and ferroptosis-associated lipid reactive oxygen species (ROS) generation resulted from knockout and overexpression of TRIB2 were all indispensible of TFRC and βTrCP. Finally, we demonstrated that TRIB2 exclusively manipulated RSL3- and erastin-induced-ferroptosis independent of GPX4 and glutathione (GSH). In conclusion, we elucidated a novel role of TRIB2 to desensitize ferroptosis via E3 βTrCP, by which facilitates TFRC ubiquitiation and finally decreases labile iron in liver cancer cells.

Skyline computation is a typical multi-objective optimization problem and has been a hot spot for current research. Most of them have concentrated on improving the efficiency of skyline computing, while the security issues associated with executing skyline computation in the cloud server are rarely taken into account. The cloud data stored in plaintext form is attacked by hackers or malicious administrators, resulting in potential leakage of users' private information. One of the most effective approaches to solve this problem is to encrypt the private data before storing it in the cloud server. However, how to compute the skyline on the encrypted data stored on a cloud server efficiently is still challenging. In this paper, we present a novel framework called SecSky. Firstly, the improved B+ tree structure and symmetric encryption were combined to construct a secure storage structure. Then, the pruning idea was applied to reduce unnecessary calculation and to achieve efficient skyline query and dynamic update. The whole process was directly completed on the ciphertext data. Finally, both the security and the efficiency of SecSky were analyzed through theoretical proof and simulation experiments. The results showed that our scheme is of sound efficiency on the premise of ensuring private data security.

Abstract Background Ferroptosis is the process of cell death triggered by lipid peroxides, and inhibition of glutathione (GSH) synthesis leads to ferroptosis. Liver cancer progression is closely linked to ferroptosis suppression. However, the mechanism by which inhibition of GSH synthesis suppresses potential ferroptosis of liver cancer cells and whether ferroptosis-related liver cancer biomarkers have a promising diagnostic value remain unknown. Methods Ribonucleotide reductase regulatory subunit M2 (RRM2) levels were measured using an enzyme linked immunosorbent assay (ELISA), quantitative RT-PCR (qPCR), immunoblotting (IB) and immunochemistry (IHC). Cell viability and cell death were measured by a CellTiter-Glo luminescent cell viability assay and staining with SYTOX Green followed by flow cytometry, respectively. Metabolites were measured using the indicated kits. The Interaction between glutathione synthetase (GSS) and RRM2 was measured using immunofluorescence (IF), co-immunoprecipitation (co-IP) and the proximal ligation assay (PLA). The diagnostic value was analyzed using the area under the receiver operating characteristic curve (AUC-ROC). Bioinformatics analysis was performed using the indicated database. Results RRM2 showed specifically elevated levels in liver cancer and inhibited ferroptosis by stimulating GSH synthesis via GSS. Mechanistically, phosphorylation of RRM2 at the Threonine 33 residue (T33) was maintained at normal levels to block the RRM2–GSS interaction and therefore protected RRM2 and GSS from further proteasome degradation. However, under ferroptotic stress, RRM2 was dephosphorylated at T33, thus the RRM2–GSS interaction was promoted. This resulted in the translocation of RRM2 and GSS to the proteasome for simultaneous degradation. Clinically, serum RRM2 was significantly associated with serum alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma glutamyl transpeptidase (γ-GT), albumin (ALB) and total bilirubin. The AUC-ROC for the combination of RRM2 with AFP was 0.947, with a sensitivity of 88.7% and a specificity of 97.0%, which indicates better diagnostic performance compared to either RRM2 or AFP alone. Conclusion RRM2 exerts an anti-ferroptotic role in liver cancer cells by sustaining GSH synthesis. Serum RRM2 will be useful as a biomarker to evaluate the degree to which ferroptosis is suppressed and improve diagnostic efficiency for liver cancer.

Abstract Background Ferroptosis, a form of regulated cell death, is an important topic in the field of cancer research. However, the signalling pathways and factors that sensitise tumour cells to ferroptosis remain elusive. Methods We determined the level of ferroptosis in cells by measuring cell death and lipid reactive oxygen species (ROS) production. The expression of RB1‐inducible coiled‐coil 1 (RB1CC1) and related proteins was analyzed by immunoblotting and immunohistochemistry. Immunofluorescence was used to determine the subcellular localization of RB1CC1. We investigated the mechanism of RB1CC1 nuclear translocation by constructing a series of RB1CC1 variants. To examine the ferroptosis‐ and RB1CC1‐dependent transcriptional program in tumour cells, chromatin immunoprecipitation sequencing was performed. To assess the effect of c‐Jun N‐terminal kinase (JNK) agonists on strenthening imidazole ketone erastin (IKE) therapy, we constructed cell‐derived xenograft mouse models. Mouse models of hepatocellular carcinoma to elucidate the importance of Rb1cc1 in IKE‐based therapy of liver tumourigenesis. Results RB1CC1 is upregulated by lipid ROS and that nuclear translocation of phosphorylation of RB1CC1 at Ser537 was essential for sensitising ferroptosis in tumour cells. Upon ferroptosis induction, nuclear RB1CC1 sharing forkhead box (FOX)‐binding motifs recruits elongator acetyltransferase complex subunit 3 (ELP3) to strengthen H4K12Ac histone modifications within enhancers linked to ferroptosis. This also stimulated transcription of ferroptosis‐associated genes, such as coiled‐coil–helix–coiled‐coil–helix domain containing 3 (CHCHD3), which enhanced mitochondrial function to elevate mitochondrial ROS early following induction of ferroptosis. FDA‐approved JNK activators reinforced RB1CC1 nuclear translocation and sensitised cells to ferroptosis, which strongly suggested that JNK is upstream of RB1CC1. Nuclear localisation of RB1CC1 correlated with lipid peroxidation in clinical lung cancer specimens. Rb1cc1 was essential for ferroptosis agonists to suppress liver tumourigenesis in mice. Conclusions Our findings indicate that RB1CC1‐associated signalling sensitises tumour cells to ferroptosis and that targeting RB1CC1 may be beneficial for tumour treatment.

Ferroptosis is an iron- and lipid peroxidation-dependent form of regulated cell death. The release of labile iron is one of the important factors affecting sensitivity to ferroptosis. Yes-associated protein (YAP) controls intracellular iron levels by affecting the transcription of ferritin heavy chain (FTH) and transferrin receptor (TFRC). However, whether YAP regulates iron metabolism through other target genes remains unknown. Here, we observed that the system Xc– inhibitor erastin inhibited the binding of the WW domain and PSY motif between YAP and transcription factor CP2 (TFCP2), and then suppressed the transcription of ferritin light chain (FTL) simultaneously mediated by YAP, TFCP2 and forkhead box A1 (FOXA1). Furthermore, inhibition of FTL expression abrogated ferroptosis-resistance in cells with sustained YAP expression. Unlike FTH, which exhibited first an increase and then a decrease in transcription, FTL transcription continued to decline after the addition of erastin, and a decrease in lysine acetyltransferase 5 (KAT5)-dependent acetylation of FTL was also observed. In lung adenocarcinoma (LUAD) tissues, lipid peroxidation and labile iron decreased, while YAP, TFCP2 and FTL increased compared to their adjacent normal tissues, and the lipid peroxidation marker 4-hydroxynonenal (4-HNE) was negatively correlated with the level of FTL or the degree of LUAD malignancy, but LUAD tissues with lower levels of 4-HNE showed a higher sensitivity to ferroptosis. In conclusion, the findings from this study indicated that the suppression of FTL transcription through the inhibition of the YAP-TFCP2-KAT5 complex could be another mechanism for elevating ferroptosis sensitivity and inducing cell death, and ferroptotic therapy is more likely to achieve better results in LUAD patients with a lower degree of lipid peroxidation.

Abstract Classic exponent-Fourier moments (EFMs) have been popularly used for image reconstruction and invariant classification. However, EFMs lack natively the translation and scaling-invariant; in addition, they exhibit two types of drawbacks, namely numerical instability and reconstruction error, which in turn influence their reconstruction capability and image classification accuracy. This study considers the challenge of defining modified EFMs (MEFMs), which are based on modified exponent polynomials. In our methods, the basis function of traditional EFMs is appropriately modified, and these modified basis functions are used to replace the original ones. The basis function of the proposed moments is composed of piecewise modified exponent polynomials modulated by a variable parameter exponential envelope. Various types of optimal-order moments can be established by slightly adjusting the bandwidth of the modified basis functions. Finally, we extend the rotation-invariant feature of previous works and propose a new method of scaling and rotation-invariant image recognition using the proposed moments in a log-polar coordinate domain. The translation invariance can then be achieved by an image projection operation, which is substituted for the traditional approach based on the calculation of image geometric moments. The experimental results demonstrate that the MEFMs perform better than traditional EFMs and other classic orthogonal moments including the latest image moments in terms of the image reconstruction capability and the invariant recognition accuracy of smoothing filters, in both noise-free and noisy conditions.

Due to the rapid development of social networks, bio-informatics, internet-of-things etc., subgraph matching query can be found in various applications. Meanwhile, the increasing popularity of storing graph data in the cloud drives demands for graph query processing on a remote cloud server. However, the query results in this scenario may not be guaranteed to be correct, especially when the cloud service provider (CSP) is malicious or compromised by some adversaries, for example, a CSP might omit some edges of the graph so that its search cost would be substantially reduced. Besides, various software bugs and unintended errors are also inevitable. All current generic verifiable computation (VC) schemes applied in this scenario are not only too impractical to be implemented but also need a lot of space to locally store their auxiliary data. To that end, we have put forth both public and designated verification schemes which focus on subgraph matching problems for outsourced graph data. They utilize a modified cryptographic primitive called accumulator to realize fast verification and low local storage overhead. In addition to the two main constructions, we have proposed an optimization to make the scheme more applicable, namely, supporting dynamic updates of the graph. At last, rigorous security proofs and efficiency analysis are given, which justify that our proposed schemes are secure and efficient, satisfying the requirements of general verifiable computation protocols.

As many services have been subcontracted to the cloud, data in such services (i.e., management applications, signal processing, and so on) are stored on cloud in encrypted form for protecting user private information. In many emerging data and signal processing applications, approximate k-nearest neighbor (ANN) query is a prerequisite component for massive high-dimensional data (such as time series, biometric, signal, multimedia data, and so on). Unfortunately, existing secure ANN (SANN) methods encounter the limitation of dimensionality. To further resolve ANN query over high-dimensionally encrypted data, in this paper, we propose an effective SANN model in Euclidean space. In overview, a secure greedy partition method is carefully designed by applying locality sensitive hashing coding and an optimized linear order. Based on that, a novel partition-based SANN (SANNP) and a multi-division version mSANNP resolve SANN query by sequentially scanning a candidate set, which is produced by matching a cloak query with a map index. Our proposed solutions guarantee security and accuracy simultaneously, and reduce communication cost significantly. Meanwhile, the greedy partition method is proved to be indistinguishable secure under chosen-plaintext attack, which is the foundation of security for the proposed solutions. Through extensive experimental studies on four data sets, the proposed mechanisms outperform the state-of-the-art approaches and provide effective and tradeoff between result accuracy and communication cost.

Abstract Background Ocular images play an essential role in ophthalmological diagnoses. Having an imbalanced dataset is an inevitable issue in automated ocular diseases diagnosis; the scarcity of positive samples always tends to result in the misdiagnosis of severe patients during the classification task. Exploring an effective computer-aided diagnostic method to deal with imbalanced ophthalmological dataset is crucial. Methods In this paper, we develop an effective cost-sensitive deep residual convolutional neural network (CS-ResCNN) classifier to diagnose ophthalmic diseases using retro-illumination images. First, the regions of interest (crystalline lens) are automatically identified via twice-applied Canny detection and Hough transformation. Then, the localized zones are fed into the CS-ResCNN to extract high-level features for subsequent use in automatic diagnosis. Second, the impacts of cost factors on the CS-ResCNN are further analyzed using a grid-search procedure to verify that our proposed system is robust and efficient. Results Qualitative analyses and quantitative experimental results demonstrate that our proposed method outperforms other conventional approaches and offers exceptional mean accuracy (92.24%), specificity (93.19%), sensitivity (89.66%) and AUC (97.11%) results. Moreover, the sensitivity of the CS-ResCNN is enhanced by over 13.6% compared to the native CNN method. Conclusion Our study provides a practical strategy for addressing imbalanced ophthalmological datasets and has the potential to be applied to other medical images. The developed and deployed CS-ResCNN could serve as computer-aided diagnosis software for ophthalmologists in clinical application.

Ocular images play an essential role in ophthalmology. Current research mainly focuses on computer-aided diagnosis using slit-lamp images, however few studies have been done to predict the progression of ophthalmic disease. Therefore exploring an effective approach of prediction can help to plan treatment strategies and to provide early warning for the patients. In this study, we present an end-to-end temporal sequence network (TempSeq-Net) to automatically predict the progression of ophthalmic disease, which includes employing convolutional neural network (CNN) to extract high-level features from consecutive slit-lamp images and applying long short term memory (LSTM) method to mine the temporal relationship of features. First, we comprehensively compare six potential combinations of CNNs and LSTM (or recurrent neural network) in terms of effectiveness and efficiency, to obtain the optimal TempSeq-Net model. Second, we analyze the impacts of sequence lengths on model's performance which help to evaluate their stability and validity and to determine the appropriate range of sequence lengths. The quantitative results demonstrated that our proposed model offers exceptional performance with mean accuracy (92.22), sensitivity (88.55), specificity (94.31) and AUC (97.18). Moreover, the model achieves real-time prediction with only 27.6ms for single sequence, and simultaneously predicts sequence data with lengths of 3-5. Our study provides a promising strategy for the progression of ophthalmic disease, and has the potential to be applied in other medical fields.

Slit-lamp images play an essential role for diagnosis of pediatric cataracts. We present a computer vision-based framework for the automatic localization and diagnosis of slit-lamp images by identifying the lens region of interest (ROI) and employing a deep learning convolutional neural network (CNN). First, three grading degrees for slit-lamp images are proposed in conjunction with three leading ophthalmologists. The lens ROI is located in an automated manner in the original image using two successive applications of Candy detection and the Hough transform, which are cropped, resized to a fixed size and used to form pediatric cataract datasets. These datasets are fed into the CNN to extract high-level features and implement automatic classification and grading. To demonstrate the performance and effectiveness of the deep features extracted in the CNN, we investigate the features combined with support vector machine (SVM) and softmax classifier and compare these with the traditional representative methods. The qualitative and quantitative experimental results demonstrate that our proposed method offers exceptional mean accuracy, sensitivity and specificity: classification (97.07%, 97.28%, and 96.83%) and a three-degree grading area (89.02%, 86.63%, and 90.75%), density (92.68%, 91.05%, and 93.94%) and location (89.28%, 82.70%, and 93.08%). Finally, we developed and deployed a potential automatic diagnostic software for ophthalmologists and patients in clinical applications to implement the validated model.