Robust mosquito species identification from diverse body and wing images using deep learning.

Mosquito-borne diseases are a major global health threat. Traditional morphological or molecular methods for identifying mosquito species often require specialized expertise or expensive laboratory equipment. The use of convolutional neural networks (CNNs) to identify mosquito species based on images may offer a promising alternative, but their practical implementation often remains limited. This study explores the applicability of CNNs in classifying mosquito species. It compares the efficacy of body and wing depictions across three image collection methods: a smartphone, macro-lens attached to a smartphone and a professional stereomicroscope. The study included 796 specimens of four morphologically similar Aedes species, Aedes aegypti, Ae. albopictus, Ae. koreicus and Ae. japonicus japonicus. The findings of this study indicate that CNN models demonstrate superior performance in wing-based classification 87.6% (95% CI: 84.2-91.0) compared to body-based classification 78.9% (95% CI: 77.7-80.0). Nevertheless, there are notable limitations of CNNs as they perform reliably across multiple devices only when trained specifically on those devices, resulting in an average decline of mean accuracy by 14%, even with extensive image augmentation. Additionally, we also estimate the required training data volume for effective classification, noting a reduced requirement for wing-based classification compared to body-based methods. Our study underscores the viability of both body and wing classification methods for mosquito species identification while emphasizing the need to address practical constraints in developing accessible classification systems.

Investigating the impact of climate and seasonality on mosquito (Diptera: Culicidae) vector populations in the connecting areas of the Tenasserim range forests in Thailand.

Mosquito-borne diseases pose a significant public health challenge globally. Our study focused on the seasonal diversity of mosquito species in the connecting areas of the Tenasserim (also known as Tanaosri) range forests in Thailand. Additionally, we employed the geometric morphometric technique to assess variations in wing size and shape among five predominant mosquito species. Throughout the study period, we collected a total of 9,522 mosquitoes, encompassing 42 species across eight genera. In these connecting areas of forests, the Simpson index and Shannon species diversity index were recorded at 0.86 and 2.36, respectively, indicating a high level of mosquito diversity. Our analysis using the Analysis of Similarities (ANOSIM) test showed significant seasonal differences in mosquito communities, with an R-value of 0.30 (p < 0.05) in the lower connecting areas and 0.37 (p < 0.05) in the upper connecting areas. Additionally, canonical correspondence analyses showed that the abundance of each mosquito species is influenced by various climate factors. Phenotypic analyses of wing size and shape have deepened our understanding of local adaptation and the seasonal pressures impacting these vectors. Notably, most species exhibited larger wing sizes in the dry season compared to other seasons. Additionally, seasonal assessments of wing shape in five predominant mosquito species revealed significant differences across seasonal populations (p < 0.05). Ongoing monitoring of these populations is crucial to enhancing our understanding of the seasonal effects on mosquito abundance and physiological adaptations. These insights are essential for developing more effective strategies to manage mosquito-borne diseases.

Morphological changes in eggs and embryos of Aedes aegypti (Diptera: Culicidae) exposed to predicted climatic scenarios for the year 2100 in the Central Amazon.

According to the IPCC, by the year 2100, rises in global temperature could reach up to 5 °C above current averages. On a planet-wide scale, this is one of the effects of climate changes that could have repercussions on the biological cycle of Aedes aegypti, the main arbovirus vector in urban environments and a transmitter of the arboviruses that cause dengue, Zika, chikungunya and urban yellow fever. The objective of this study was to evaluate morphological changes in Ae. aegypti eggs and embryos maintained in a climate change simulator. For this, specimens obtained from an insectarium were kept in four chambers that simulated the range of environmental scenarios predicted by the IPCC for the year 2100. The eggs obtained from each room were collected and transported to the laboratory for morphometric and morphological analysis, using confocal and scanning microscopy. Aedes aegypti eggs (n=20) were used to obtain the following variables: total width, total length, length-width ratio and diameter of the micropylar disc. Additionally, 20 embryos were used to obtain the data on head capsule length, width and length-width ratio. The data were subjected to a normality test and the means of each variable were compared using ANOVA and Tukey's post-hoc test, considering (p ≤ 0.05). A significant reduction (p < 0.05) was observed mainly in the mean lengths under the current-extreme scenario (587.5 and 553.6 µm, respectively), as well as in the widths under the current-mild scenario (171 and 158.4 µm, respectively). The length of the cephalic capsule was also affected, showing significant differences in the means under the current-intermediate scenario (189.5 and 208.5 µm, respectively), as well as in the widths between the current-intermediate scenarios (173.7 and 194.9 µm, respectively). The results suggest significant changes in the morphometry of Ae. aegypti eggs and embryos as a result of the climatic influences to which the adults were subjected, which may have an impact on vector population density and, consequently, on arbovirus dynamics in urban environments.

Resolution of cryptic mosquito species through wing morphometrics.

Mosquitoes are medically important insects, and accurate species identification is crucial to understanding vector biology, forming the cornerstone of successful vector control programs. Identification is difficult owing to morphologically similar species. Wing morphometrics can provide a simple, fast, and accurate way to classify species, and using it as a method to differentiate vector species among its cryptic congeners has been underexplored. Using a total of 227 mosquitoes and 20 landmarks per specimen, we demonstrated the utility of wing morphometrics in differentiating species two groups occurring in sympatry - Culex (Culex) vishnui group and Culex (Lophoceraomyia) subgenus, as well as explored population-level variation in the wing shape of Aedes albopictus across habitats. Cytochrome oxidase subunit I (COI) gene region was sequenced to validate the morphological and morphometric identification. Procrustes ANOVA regression and CVA based on wing shape reflected that the wing landmarks across all species differed significantly, and leave-one-out cross validation revealed an overall high accuracy of >97% for the two Culex groups. Wing morphometrics uncovered population-level variation within Aedes albopictus, but cross validation accuracy was low. Overall, we show that wing geomorphometric analysis is able to resolve cryptic Culex species (including vectors) occurring sympatrically, and is a robust tool for identifying mosquitoes reliably.

Description of New Morphological Variation of Culex (Culex) coronator Dyar and Knab, 1906 and First Report of Culex (Carrollia) bonnei Dyar, 1921 Found in the Central Region of Peru.

Mosquitoes (Diptera: Culicidae) pose a significant threat to public health worldwide, especially in tropical and subtropical regions, where they act as primary vectors in transmission of infectious agents. In Peru, 182 culicid species have been identified and several species of the genus Culex are known to transmit arboviruses. However, knowledge of mosquito diversity and distribution remains limited, with many studies focusing on specific regions only. Here, we describe a new morphological variation of Cx. (Culex) coronator Dyar and Knab, 1906, and report the presence of Culex (Carrollia) bonnei Dyar, 1921 in the central region of Peru, Huanuco. Specimens were obtained through larvae collections and identified through morphologic characterization, including dissection of male genitalia, and molecular analyses. In total, 17 mosquitoes were analyzed, and the genitalia of the male specimens allowed the identification of Cx. coronator and Cx. bonnei. Partial sequences of the CoxI gene corresponding to these two species were obtained (N = 10). Phylogenetic analysis revealed that the sequences of Cx. coronator grouped in a monophyletic clade with sequences ascribed to other species corresponding to the subgenus Carrollia, while Cx. bonnei specimens formed a monophyletic clade with homologous sequences from GenBank. This study underscores the importance of continued efforts to study the diversity and distribution of mosquitoes in Peru, including their potential role as vectors of human pathogens, to underpin effective disease control and prevention strategies, highlighting the importance of a complemented morphological and molecular analysis.

A DNA barcode reference library of Croatian mosquitoes (Diptera: Culicidae): implications for identification and delimitation of species, with notes on the distribution of potential vector species.

BACKGROUND: Mosquitoes pose a risk to human health worldwide, and correct species identification and detection of cryptic species are the most important keys for surveillance and control of mosquito vectors. In addition to traditional identification based on morphology, DNA barcoding has recently been widely used as a complementary tool for reliable identification of mosquito species. The main objective of this study was to create a reference DNA barcode library for the Croatian mosquito fauna, which should contribute to more accurate and faster identification of species, including cryptic species, and recognition of relevant vector species. METHODS: Sampling was carried out in three biogeographical regions of Croatia over six years (2017-2022). The mosquitoes were morphologically identified; molecular identification was based on the standard barcoding region of the mitochondrial COI gene and the nuclear ITS2 region, the latter to identify species within the Anopheles maculipennis complex. The BIN-RESL algorithm assigned the COI sequences to the corresponding BINs (Barcode Index Number clusters) in BOLD, i.e. to putative MOTUs (Molecular Operational Taxonomic Units). The bPTP and ASAP species delimitation methods were applied to the genus datasets in order to verify/confirm the assignment of specimens to specific MOTUs. RESULTS: A total of 405 mosquito specimens belonging to six genera and 30 morphospecies were collected and processed. Species delimitation methods assigned the samples to 31 (BIN-RESL), 30 (bPTP) and 28 (ASAP) MOTUs, with most delimited MOTUs matching the morphological identification. Some species of the genera Culex, Aedes and Anopheles were assigned to the same MOTUs, especially species that are difficult to distinguish morphologically and/or represent species complexes. In total, COI barcode sequences for 34 mosquito species and ITS2 sequences for three species of the genus Anopheles were added to the mosquito sequence database for Croatia, including one individual from the Intrudens Group, which represents a new record for the Croatian mosquito fauna. CONCLUSION: We present the results of the first comprehensive study combining morphological and molecular identification of most mosquito species present in Croatia, including several invasive and vector species. With the exception of some closely related species, this study confirmed that DNA barcoding based on COI provides a reliable basis for the identification of mosquito species in Croatia.

The monsoon-associated equine South African pointy mosquito 'Aedes caballus'; the first comprehensive record from southeastern Iran with a description of ecological, morphological, and molecular aspects.

The equine South African pointy vector mosquito, Aedes caballus, poses a significant threat to human health due to its capacity for transmitting arboviruses. Despite favorable climate for its existence in southeast Iran, previous records of this species in the area have indicated very low abundance. This comprehensive field and laboratory study aimed to assess its current adult population status in this region, utilizing a combination of ecological, morphological and molecular techniques. Four distinct types of traps were strategically placed in three fixed and two variable mosquito sampling sites in the southern strip of Sistan and Baluchistan Province. Subsequently, DNA was extracted from trapped mosquitoes and subjected to PCR amplification using the molecular markers COI, ITS2, and ANT. In total, 1734 adult Ae. caballus specimens were collected from rural areas, with the majority being captured by CO2-baited bednet traps. A notable increase in the abundance of this species was observed following rainfall in February. The genetic analysis revealed multiple haplotypes based on COI and ITS2 sequences, with COI gene divergence at 0.89%, and ITS2 sequence divergence at 1.6%. This suggests that previous challenges in morphological identification may have led to misidentifications, with many adults previously classified as Ae. vexans potentially being Ae. caballus. The findings of this study hold significant implications for public health authorities, providing valuable insights for integrated and targeted vector control and disease management efforts.

Demographic inference from the mt-DNA COI gene and wing geometry of Culex gelidus (Diptera: Culicidae), an important vector of Japanese encephalitis in Thailand.

Culex gelidus (Diptera: Culicidae), an important vector of the Japanese encephalitis virus (JEV), contributes to human viral encephalitis in many Asian countries, including Thailand. This study represents the first investigation of the demographic patterns of Cx. gelidus populations in Thailand using cytochrome c oxidase subunit I (COI) gene analysis and wing geometric morphometrics (GM). Mosquitoes were collected from 10 provinces across six regions of Thailand in 2022. Analysis of the COI sequences (n = 182) indicated high haplotype diversity (0.882) and low nucleotide diversity (0.006), with 72 haplotypes identified. The haplotype network demonstrated no profound splits among the geographic populations. Neutral tests, including Tajima's D and Fu's Fs, displayed negative values, with a significant result observed for Fu's Fs (-33.048, p < 0.05). The mismatch distribution analysis indicated that the population does not statistically deviate from a model of sudden population expansion (SSD = 0.010, p > 0.05; Rg = 0.022, p > 0.05). The estimations suggest that the Cx. gelidus population in Thailand began its expansion approximately between 459,243 and 707,011 years ago. The Mantel test showed no significant relationship between genetic and geographic distances (r = 0.048, p > 0.05). Significant phenotypic differences (based on wing shape) were observed among most populations. Additionally, in this study, we found no significant relationships between phenotypic and genetic distances (r = 0.250, p > 0.05). Understanding the genetic and morphological dynamics of Cx. gelidus is vital for developing targeted surveillance and vector control measures. This knowledge will also help to predict how future environmental changes might affect these populations, thereby informing long-term vector management strategies.

Influence of insular conditions on wing phenotypic variation in two dominant mosquito vectors, Aedes albopictus and Armigeres subalbatus (Diptera: Culicidae), in the border archipelagos of Thailand.

Insects geographically separated into island and mainland populations often exhibit phenotypic variations, a phenomenon known as insular conditions. These conditions can lead to rapid evolutionary changes that affect the morphological characteristics of mosquito vectors. Nevertheless, studies that specifically examine phenotype differences between island and mainland mosquito populations have been limited. In this study, wing variation in size and shape was investigated using the geometric morphometric (GM) technique in two dominant mosquito vectors, Aedes albopictus and Armigeres subalbatus, in the Ranong and Trat archipelagos of Thailand. Significant differences in average wing centroid size (CS) were found in 6 out of 15 population pairs for Ae. albopictus (p < 0.05) and in 5 population pairs for Ar. subalbatus (p < 0.05). After removing the allometric effect, canonical variate analyses (CVA) based on wing shape analysis revealed overlap across all populations for both Ae. albopictus and Ar. subalbatus. However, the statistical analysis indicated that Ar. subalbatus exhibited wing shape differences across all populations (p < 0.05), and most Ae. albopictus populations also displayed distinct wing shapes (p < 0.05), except for the populations from Chang Island and the mainland of Ranong, which showed no significant differences (p > 0.05). These findings enhance our understanding of mosquito adaptability in island regions and provide valuable data for the surveillance and monitoring of vector evolution.

DNA Barcoding Analysis of Trinidad Haemagogus Mosquitoes Reveals Evidence for Putative New Species.

Background: Haemagogus janthinomys is a primary sylvan vector of yellow fever virus and the emerging Mayaro virus. However, despite its medical importance, there is a dearth of data on the molecular taxonomy of this mosquito species. Methods: In this study, DNA barcoding analysis was performed on 64 adult female mosquitoes from Trinidad morphologically identified as Hg. janthinomys. The mitochondrial cytochrome c oxidase I (COI) gene and ribosomal DNA internal transcribed spacer 2 (ITS2) region of the mosquitoes were PCR amplified and sequenced, and molecular phylogenies inferred. Results: The BLASTN analysis showed that only 20% (n = 13/66) of COI sequences had high similarity (>99% identity) to Hg. janthinomys and the remaining sequences had low similarity (<90% identity) to reference GenBank sequences. Phylogenetic analysis of COI sequences revealed the presence of four strongly supported groups, with one distinct clade that did not align with any reference sequences. Corresponding ITS2 sequences for samples in this distinct COI group clustered into three clades. Conclusions: These molecular findings suggest the existence of a putative new Haemagogus mosquito species and underscore the need for further, more in-depth investigations into the taxonomy and classification of the Haemagogus genus.

Classification and identification of mosquitoes in China based on rDNA 28S D5 region.

Accurate classification and identification of mosquitoes are essential for the prevention and control of mosquito-borne diseases. In this study, adult mosquitoes were collected from 15 cities across 14 provinces in China. They were identified morphologically with the dominant species determined. Furthermore, representative samples were identified at the molecular level based on rDNA 28S D5. In total, 880 adult mosquitoes were collected belonging to Culex (266), Aedes (473), Armigeres (13), and Anopheles (5). Aedes albopictus and "C. pipiens subgroup" were the dominant species. A total of 140 sequences of 28S D5 region (68 for "C. pipiens subgroup", 51 for Ae. albopictus, 18 for Ar. subalbatus, and three for An. sinensis) ranging from 148 to 161 bp were obtained, with 100 % success of amplification and sequencing. Molecular identification were consistent with morphological classification. Sequence analysis showed that "C. pipiens subgroup" was identified into three clades: the traditional C. pipiens subgroup (Clade I), the newly discovered C. cf. perexiguus (Clade II), and C. new sp. (Clade III). Clade I contained the most abundant haplotypes (16) widely distributed without geographical differences. Clade II included six haplotypes that were aggregately distributed south of the Yangtze River. Only three sequences in Clade III showed two haplotypes with no geographical differences. Further morphological comparisons demonstrated differences in body color, beaks, and abdomens among the three clades. In conclusion, the rDNA 28S D5 region could effectively distinguish Culex, Aedes, Armigeres, and Anopheles species at the lower category level, demonstrating its potential as a mini-DNA barcode for mosquito identification.

Molecular Identification and phylogenetic analysis of mosquito vectors from Mananthavady Taluk, Wayanad, Kerala, India.

BACKGROUND & OBJECTIVES: Every year more than one billion people are infected and about one million people die from vector-borne diseases; of which mosquito-borne diseases remain as the world's most severe insect-borne diseases with excessive rates of morbidity and mortality. This study aimed to examine the mosquito vectors and the possible diseases transmitted by them in the Mananthavady Taluk of Wayanad, Kerala. METHODS: The area selected for the present study was Mananthavady Taluk of Wayanad district, Kerala, during 2019-2021. The collected specimen were subjected for morphological identification using taxonomic keys and were confirmed by DNA barcoding. Molecular phylogeny assessment was done for the collected species of vector mosquitoes. RESULTS: A total of 17 mosquito species belonging to 5 genera, Anopheles, Aedes, Culex, Mansonia and Armigereswere identified. The mitochondrial COI gene sequences generated for molecular identification of these species were submitted to NCBI GenBank. INTERPRETATION & CONCLUSION: Overall, this study extends our understanding of the molecular evolution of mosquito vectors of medical and veterinary concern, which could aid in developing biotechnological approaches used in Culicidae control programs.

Genetic and morphological polymorphisms of Aedes scapularis (Diptera: Culicidae), vector of filariae and arboviruses.

BACKGROUND: Aedes scapularis is a neotropical mosquito that is competent to vector viruses and filariae. It is reputed to be highly morphologically and genetically polymorphic, facts that have raised questions about whether it is a single taxonomic entity. In the last five decades, authors have posed the hypothesis that it could actually be a species complex under incipient speciation. Due to its epidemiological importance, its taxonomic status should be determined with confidence. AIM AND METHOD: Our objective was to investigate more deeply the polymorphism of Ae. scapularis to detect any evidence of incipient speciation of cryptic species. We then compared populational samples from the Southeastern, Northern and Northeastern regions of Brazil. The biological markers used in the comparison were: the complete mitochondrial DNA, the isolated mitochondrial gene cytochrome oxidase subunit I (COI) and wing geometry. RESULTS AND DISCUSSION: As expected, high morphological/genetic polymorphism was observed in all Ae. scapularis populations, however it was not indicative of segregation or incipient speciation. There was no correlation between wing shape and the geographical origin of the populations analysed. A congruent observation resulted from the analysis of the COI gene, which revealed a high number of haplotypes (51) and no clusterization of populational samples according to the original biomes. In the phylogenetic analysis of the 13 mitochondrial protein-coding genes, the Ae. scapularis clade clustered with maximum support (100% bootstrap support and posterior probability of 1). No significant internal structure was observed in the Ae. scapularis clade, which was nearly a polytomy. Taken together, our results indicate that this species is not a species complex. CONCLUSION: We conclude that there was no indication, in the analysed regions, of the occurrence of more than one taxon in the species Ae. scapularis, despite it being highly polymorphic. By ruling out the former species complex hypothesis, our phylogenetic results reinforce that Ae. scapularis is a single taxonomic unit and should be monitored with standardized surveillance and control methods.

First Record of Natural Transovarial Transmission of Dengue Virus in Aedes albopictus from Cuba.

Transovarial transmission (TOT) of dengue virus (DENV) in Aedes spp. is an important mechanism for DENV maintenance in nature and may be important in initiating outbreaks. The objective of this study was to explore the occurrence of TOT in wild Aedes albopictus populations in Cuba. Mosquito larvae were collected in Cotorro municipality, Havana, Cuba, and identified to species. Fifteen pools of Ae. albopictus each containing 30 larvae were processed for DENV detection by using conventional reverse transcription polymerase chain reaction (RT-PCR) and nested PCR. Four out of 15 pools processed were positive for DENV-3, but no other DENV serotype was detected. This is the first time TOT of DENV detected in Cuban field populations of Ae. albopictus, and this suggests that this species may be an important vector of DENV in Cuba.

Geometric morphometric wing analysis as a tool to discriminate female mosquitoes from different suburban areas of Chiang Mai province, Thailand.

Mosquitoes are hematophagous insects that transmit parasites and pathogens with devastating effects on humans, particularly in subtropical regions. Different mosquito species display various behaviors, breeding sites, and geographic distribution; however, they can be difficult to distinguish in the field due to morphological similarities between species and damage caused during trapping and transportation. Vector control methods for controlling mosquito-borne disease epidemics require an understanding of which vector species are present in the area as well as the epidemiological patterns of disease transmission. Although molecular techniques can accurately distinguish between mosquito species, they are costly and laborious, making them unsuitable for extensive use in the field. Thus, alternative techniques are required. Geometric morphometrics (GM) is a rapid and inexpensive technique that can be used to analyze the size, shape, and shape variation of individuals based on a range of traits. Here, we used GM to analyze the wings of 1,040 female mosquitoes from 12 different species in Thailand. The right wing of each specimen was removed, imaged microscopically, and digitized using 17 landmarks. Wing shape variation among genera and species was analyzed using canonical variate analysis (CVA), while discriminant function analysis was used to cross-validate classification reliability based on Mahalanobis distances. Phenetic relationships were constructed to illustrate the discrimination patterns for genera and species. CVA of the morphological variation among Aedes, Anopheles, Armigeres, Culex, and Mansonia mosquito genera revealed five clusters. In particular, we demonstrated a high percentage of correctly-distinguished samples among Aedes (97.48%), Armigeres (96.15%), Culex (90.07%), and Mansonia (91.67%), but not Anopheles (64.54%). Together, these findings suggest that wing landmark-based GM analysis is an efficient method for identifying mosquito species, particularly among the Aedes, Armigeres, Culex, and Mansonia genera.

Modeling the effects of Aedes aegypti's larval environment on adult body mass at emergence.

Mosquitoes vector harmful pathogens that infect millions of people every year, and developing approaches to effectively control mosquitoes is a topic of great interest. However, the success of many control measures is highly dependent upon ecological, physiological, and life history traits of mosquito species. The behavior of mosquitoes and their potential to vector pathogens can also be impacted by these traits. One trait of interest is mosquito body mass, which depends upon many factors associated with the environment in which juvenile mosquitoes develop. Our experiments examined the impact of larval density on the body mass of Aedes aegypti mosquitoes, which are important vectors of dengue, Zika, yellow fever, and other pathogens. To investigate the interactions between the larval environment and mosquito body mass, we built a discrete time mathematical model that incorporates body mass, larval density, and food availability and fit the model to our experimental data. We considered three categories of model complexity informed by data, and selected the best model within each category using Akaike's Information Criterion. We found that the larval environment is an important determinant of the body mass of mosquitoes upon emergence. Furthermore, we found that larval density has greater impact on body mass of adults at emergence than on development time, and that inclusion of density dependence in the survival of female aquatic stages in models is important. We discuss the implications of our results for the control of Aedes mosquitoes and on their potential to spread disease.

Evidence for the spread of the alien species Aedes koreicus in the Lombardy region, Italy.

BACKGROUND: Aedes koreicus is a mosquito species characterized by marked anthropophilic behavior, and a potential vector of nematodes and viruses. It is native to East Asia, but its presence has recently been reported in many regions of Europe. In Italy, these mosquitoes had been detected in the northeast since 2011 and are now spreading towards the southwest of the country. METHODS: In 2020, during a surveillance program for invasive mosquito species in the district of Bergamo (Lombardy Region, Italy), about 6000 mosquito larvae were collected. Emerged adults were assigned to mosquito species according to morphological analyses, followed by amplification and sequencing of genetic markers (COI, ND4, ITS2 and D2). RESULTS: According to the morphological and genetic data, about 50 individuals belonged to the species Ae. koreicus. CONCLUSION: We report the presence of Ae. koreicus in the district of Bergamo, which confirms the spread of this species in the north of Italy and raises concerns about its possible role as a vector of diseases in the Alpine area.

Geometric morphometrics and machine learning as tools for the identification of sibling mosquito species of the Maculipennis complex (Anopheles).

Geometric morphometrics allows researchers to use the specific software to quantify and to visualize morphological differences between taxa from insect wings. Our objective was to assess wing geometry to distinguish four Anopheles sibling species of the Maculipennis complex, An. maculipennis s. s., An. daciae sp. inq., An. atroparvus and An. melanoon, found in Northern Italy. We combined the geometric morphometric approach with different machine learning alghorithms: support vector machine (SVM), random forest (RF), artificial neural network (ANN) and an ensemble model (EN). Centroid size was smaller in An. atroparvus than in An. maculipennis s. s. and An. daciae sp. inq. Principal component analysis (PCA) explained only 33% of the total variance and appeared not very useful to discriminate among species, and in particular between An. maculipennis s. s. and An. daciae sp. inq. The performance of four different machine learning alghorithms using procrustes coordinates of wing shape as predictors was evaluated. All models showed ROC-AUC and PRC-AUC values that were higher than the random classifier but the SVM algorithm maximized the most metrics on the test set. The SVM algorithm with radial basis function allowed the correct classification of 83% of An. maculipennis s. s. and 79% of An. daciae sp. inq. ROC-AUC analysis showed that three landmarks, 11, 16 and 15, were the most important procrustes coordinates in mean wing shape comparison between An. maculipennis s. s. and An. daciae sp. inq. The pattern in the three-dimensional space of the most important procrustes coordinates showed a clearer differentiation between the two species than the PCA. Our study demonstrated that machine learning algorithms could be a useful tool combined with the wing geometric morphometric approach.

Mosquito species identification using convolutional neural networks with a multitiered ensemble model for novel species detection.

With over 3500 mosquito species described, accurate species identification of the few implicated in disease transmission is critical to mosquito borne disease mitigation. Yet this task is hindered by limited global taxonomic expertise and specimen damage consistent across common capture methods. Convolutional neural networks (CNNs) are promising with limited sets of species, but image database requirements restrict practical implementation. Using an image database of 2696 specimens from 67 mosquito species, we address the practical open-set problem with a detection algorithm for novel species. Closed-set classification of 16 known species achieved 97.04 ± 0.87% accuracy independently, and 89.07 ± 5.58% when cascaded with novelty detection. Closed-set classification of 39 species produces a macro F1-score of 86.07 ± 1.81%. This demonstrates an accurate, scalable, and practical computer vision solution to identify wild-caught mosquitoes for implementation in biosurveillance and targeted vector control programs, without the need for extensive image database development for each new target region.

Implementation of a deep learning model for automated classification of Aedes aegypti (Linnaeus) and Aedes albopictus (Skuse) in real time.

Classification of Aedes aegypti (Linnaeus) and Aedes albopictus (Skuse) by humans remains challenging. We proposed a highly accessible method to develop a deep learning (DL) model and implement the model for mosquito image classification by using hardware that could regulate the development process. In particular, we constructed a dataset with 4120 images of Aedes mosquitoes that were older than 12 days old and had common morphological features that disappeared, and we illustrated how to set up supervised deep convolutional neural networks (DCNNs) with hyperparameter adjustment. The model application was first conducted by deploying the model externally in real time on three different generations of mosquitoes, and the accuracy was compared with human expert performance. Our results showed that both the learning rate and epochs significantly affected the accuracy, and the best-performing hyperparameters achieved an accuracy of more than 98% at classifying mosquitoes, which showed no significant difference from human-level performance. We demonstrated the feasibility of the method to construct a model with the DCNN when deployed externally on mosquitoes in real time.