Tongue Image Research Articles

µTongue: A Microfluidics-Based Functional Imaging Platform for the Tongue In Vivo.

Intravital fluorescence microscopy is a tool used widely to study multicellular dynamics in a live animal. However, it has not been successfully used in the taste sensory organ. By integrating microfluidics into the intravital tongue imaging window, the µTongue provides reliable functional images of taste cells in vivo under controlled exposure to multiple tastants. In this paper, a detailed step-by-step procedure to utilize the µTongue system is presented. There are five subsections: preparing of tastant solutions, setting up of a microfluidic module, sample mounting, acquiring functional image data, and data analysis. Some tips and techniques to solve the practical issues that may arise when using the µTongue are also presented.

Lenition and fortition of /r/ in utterance-final position, an ultrasound tongue imaging study of lingual gesture timing in spontaneous speech

The most fundamental division in English dialects is the rhotic/non-rhotic division. The mechanisms of historical /r/-loss sound change are not well understood, but studying a contemporary /r/-loss sound change in a rhotic variety of English can provide new insights. We know that /r/ weakening in contemporary Scottish English is a gesture-timing based phenomenon and that it is socially indexical, but we have no phonetic explanation for the predominance of weak /r/ variants in utterance-final position. Using a socially-stratified conversational ultrasound tongue imaging speech corpus, this study investigates the effects of boundary context, along with other linguistic and social factors such as syllable stress, following-consonant place and social class, on lingual gesture timing in /r/ and strength of rhoticity. Mixed-effects modelling identified that utterance-final context conditions greater anterior lingual gesture delay in /r/ and weaker-sounding /r/s, but only in working-class speech. Middle-class speech shows no anterior lingual gesture delay for /r/ in utterance-final position and /r/ is audibly strengthened in this position. It is unclear whether this divergence is due to variation in underlying tongue shape for /r/ in these social-class communities, or whether utterance-final position provides a key location for the performance of social class using salient variants of /r/.

Design and Simulation of Ear and Tongue Based Biometrics for Attendance Management System

Multimodal biometrics has engrossed a lot of interest in recent years as it offers a more dependable scheme for verification and authentication. Multimodal biometrics includes a combination of information from different biometric modalities. This research work simulates a novel method with a view of attendance management using ear and tongue images. Principal Component Analysis was employed for feature extraction of the biometric modalities and Self Organizing Feature Map was used for training and testing of the system. The method was evaluated using six thousand ear images and four thousand tongue images. The acquired images were preprocessed and features extracted were fused at this stage. The fusion results of ear and tongue images demonstrated better performance for attendance management.

A weakly supervised tooth‐mark and crack detection method in tongue image

AbstractTongue diagnosis is one of the primary clinical diagnostic methods in Traditional Chinese Medicine. Recognizing the tooth‐marked tongue and the crackled tongue plays an essential role in evaluating the status of patients. Previous methods mainly focus on identifying whether a tongue image is a tooth‐marked tongue (cracked tongue) or not, while cannot provide more details. In this study, we propose a weakly supervised method for training the tooth‐mark and crack detection model by leveraging fully bounding‐box level annotated and coarse image‐level annotated tongue images. The proposed model is extended from the YOLO object detection model, and we add several classification branches for recognizing the tooth‐marked tongue and cracked tongue. The classification branch aims to predict the coarse label for both coarse‐labeled data and fully annotated data. The detection branch is used to locate the position of tooth marks and cracks from the fully annotated data. Finally, we utilize a multitask loss function for training the model. Experimental results on a challenging tongue image dataset demonstrate the effectiveness of our proposed weakly supervised method.

The Dorsal Differentiation of Velar From Alveolar Stops in Typically Developing Children and Children With Persistent Velar Fronting.

Purpose This study has two key aims: first, to provide developmental articulatory norms for the alveolar-velar distinction in 30 English-speaking typically developing (TD) children; second, to illustrate the utility of the reported measures for classifying and quantifying the speech of children with a history of persistent velar fronting as they develop the contrast longitudinally. Method This study involved secondary data analysis of the UltraSuite corpus comprising ultrasound tongue imaging recordings of speech materials from 30 typical children and longitudinal data from five children with persistent velar fronting undergoing ultrasound visual biofeedback intervention. We present two new measures of coronal dorsal differentiation: KTMax and KT crescent area. These measures distinguish /k/ and /t/ by quantifying the magnitude of this distinction in absolute spatial terms (mm of linear dorsal difference). For the typical children, we report these measures in corner vowel contexts. We then compare these to dorsal productions by the children with speech disorders, before, during, and after intervention. Results Both measures reliably distinguished /k/ and /t/ in TD children. There was an effect of vowel, with larger KTmax and KT crescent area in /a/ and /o/ vowel contexts than in an /i/ context. The children with persistent velar fronting showed KTmax values near zero before intervention, showing a complete merger between /k/ and /t/. During intervention, they showed variable KTmax values. Post intervention, they showed values within the range of typical children. Conclusions This study provides articulatory norms derived from ultrasound tongue imaging for the dorsal differentiation in alveolar and velar stops in TD children. By applying these norms to children with persistent velar fronting as they acquire this contrast, we see that /k/ is acquired in an articulatorily gradient manner.

The tongue features associated with chronic kidney disease.

Background:Traditional Chinese medicine (TCM) tongue diagnosis plays an important role in differentiation of symptoms because the tongue reflects the physiological and pathological condition of the body. The automatic tongue diagnosis system (ATDS), which noninvasively captures tongue images, can provide objective and reliable diagnostic information. Chronic kidney disease (CKD) currently is an important global public health problem and contributor to morbidity and mortality from non-communicable diseases. Thus, it is interesting to analyze and probe the relationship between tongue examination and CKD.Methods:This protocol is a cross-sectional, case-controlled observational study investigating the usefulness of the ATDS in clinical practice by examining its efficacy as a diagnostic tool for CKD. Volunteers over 20 years old with and without CKD will be enrolled. Tongue images will be captured and the patients divided into 2 groups: CKD group and healthy group. Nine primary tongue features will be extracted and analyzed, including tongue shape, tongue color, tooth mark, tongue fissure, fur color, fur thickness, saliva, ecchymosis, and red dots.Result:The results of this study will systematically evaluate tongue manifestations of patients and examine its efficacy as an early detection and diagnosis of CKD.Discussion:The aim of this protocol is to investigate discriminating tongue features to distinguish between CKD and normal people, and establish differentiating index to facilitate the noninvasive detection of CKD.Trial registries:ClinicalTrials.gov; Identifier: NCT04708743.

Prospective motion detection and re-acquisition in diffusion MRI using a phase image-based method-Application to brain and tongue imaging.

To develop an image-based motion-robust diffusion MRI (dMRI) acquisition framework that is able to minimize motion artifacts caused by rigid and nonrigid motion, applicable to both brain and tongue dMRI. We developed a novel prospective motion-correction technique in dMRI using a phase image-based real-time motion-detection method (PITA-MDD) with re-acquisition of motion-corrupted images. The prospective PITA-MDD acquisition technique was tested in the brains and tongues of volunteers. The subjects were instructed to move their heads or swallow, to induce motion. Motion-detection efficacy was validated against visual inspection as the gold standard. The effect of the PITA-MDD technique on diffusion-parameter estimates was evaluated by comparing reconstructed fiber tracts using tractography with and without re-acquisition. The prospective PITA-MDD technique was able to effectively and accurately detect motion-corrupted data as compared with visual inspection. Tractography results demonstrated that PITA-MDD motion detection followed by re-acquisition helps in recovering lost and misshaped fiber tracts in the brain and tongue that would otherwise be corrupted by motion and yield erroneous estimates of the diffusion tensor. A prospective PITA-MDD technique was developed for dMRI acquisition, providing improved dMRI image quality and motion-robust diffusion estimation of the brain and tongue.

An Unsupervised Tongue Segmentation Algorithm Based on Improved gPb-owt-ucm

Background and Objective: The modernization of tongue diagnosis is an important research in Traditional Chinese Medicine. Accurate and practical tongue segmentation method is a premise in subsequent analyses. In this paper, an unsupervised tongue segmentation method is proposed based on an improved gPb-owt-ucm algorithm. The gPb-owt-ucm is short for global pixel point, oriented watershed transform and ultrametric contour map. Methods: Improved gPb-owt-ucm algorithm is adopted in this paper because of its powerful contour detection capabilities. The boundary feasibility of each pixel is calculated by the weight of pixel, and the result is converted to multiple closed regions and hierarchical tree. Finally, locating tongue accurate boundary by rectangular slider is taken to perform the final tongue segmentation. Two experiments are designed to evaluate its effectiveness by comparing with the snake method. Results: 300 tongue images were tested (150 images for the diabetes and 150 images for the health) in two experiments. The first one is to validate boundary detection performance (CBDR experiment). The second one is for validation of classification performance (CCE experiment) between diabetic and healthy tongues. In CBDR experiment, the mean and variance of IoU obtained using our improved gPb-owt-ucm method are 0.72±0.19, which are better than the snake method. In CCE experiment, the obtained precision and F1-score using our method are 1.0 and 0.97 over diabetic data respectively, and results of 0.94, 0.97 over health data. Conclusion: The effectiveness of our improved unsupervised gPb-owt-ucm method is validated in comparisons with the snake method. In the future, we plan to combine the proposed method with a supervised method in order to achieve more improvements for the tongue segmentation.

Exploiting ultrasound tongue imaging for the automatic detection of speech articulation errors

Speech sound disorders are a common communication impairment in childhood. Because speech disorders can negatively affect the lives and the development of children, clinical intervention is often recommended. To help with diagnosis and treatment, clinicians use instrumented methods such as spectrograms or ultrasound tongue imaging to analyse speech articulations. Analysis with these methods can be laborious for clinicians, therefore there is growing interest in its automation. In this paper, we investigate the contribution of ultrasound tongue imaging for the automatic detection of speech articulation errors. Our systems are trained on typically developing child speech and augmented with a database of adult speech using audio and ultrasound. Evaluation on typically developing speech indicates that pre-training on adult speech and jointly using ultrasound and audio gives the best results with an accuracy of 86.9%. To evaluate on disordered speech, we collect pronunciation scores from experienced speech and language therapists, focusing on cases of velar fronting and gliding of /r/. The scores show good inter-annotator agreement for velar fronting, but not for gliding errors. For automatic velar fronting error detection, the best results are obtained when jointly using ultrasound and audio. The best system correctly detects 86.6% of the errors identified by experienced clinicians. Out of all the segments identified as errors by the best system, 73.2% match errors identified by clinicians. Results on automatic gliding detection are harder to interpret due to poor inter-annotator agreement, but appear promising. Overall findings suggest that automatic detection of speech articulation errors has potential to be integrated into ultrasound intervention software for automatically quantifying progress during speech therapy.

A tongue features fusion approach to predicting prediabetes and diabetes with machine learning

BackgroundDiabetics has become a serious public health burden in China. Multiple complications appear with the progression of diabetics pose a serious threat to the quality of human life and health. We can prevent the progression of prediabetics to diabetics and delay the progression to diabetics by early identification of diabetics and prediabetics and timely intervention, which have positive significance for improving public health. ObjectiveUsing machine learning techniques, we establish the noninvasive diabetics risk prediction model based on tongue features fusion and predict the risk of prediabetics and diabetics. MethodsApplying the type TFDA-1 Tongue Diagnosis Instrument, we collect tongue images, extract tongue features including color and texture features using TDAS, and extract the advanced tongue features with ResNet-50, achieve the fusion of the two features with GA_XGBT, finally establish the noninvasive diabetics risk prediction model and evaluate the performance of testing effectiveness. ResultsCross-validation suggests the best performance of GA_XGBT model with fusion features, whose average CA is 0.821, the average AUROC is 0.924, the average AUPRC is 0.856, the average Precision is 0.834, the average Recall is 0.822, the average F1-score is 0.813. Test set suggests the best testing performance of GA_XGBT model, whose average CA is 0.81, the average AUROC is 0.918, the average AUPRC is 0.839, the average Precision is 0.821, the average Recall is 0.81, the average F1-score is 0.796. When we test prediabetics with GA_XGBT model, we find that the AUROC is 0.914, the Precision is 0.69, the Recall is 0.952, the F1-score is 0.8. When we test diabetics with GA_XGBT model, we find that the AUROC is 0.984, the Precision is 0.929, the Recall is 0.951, the F1-score is 0.94. ConclusionsBased on tongue features, the study uses classical machine learning algorithm and deep learning algorithm to maximum the respective advantages. We combine the prior knowledge and potential features together, establish the noninvasive diabetics risk prediction model with features fusion algorithm, and detect prediabetics and diabetics noninvasively. Our study presents a feasible method for establishing the association between diabetics and the tongue image information and prove that tongue image information is a potential marker which facilitates effective early diagnosis of prediabetics and diabetics.

Recommending prescription via tongue image to assist clinician

Traditional Chinese Medicine often use the prescription composed of herbs to cure the disease, which requires doctors with the rich professional knowledge and experience. It is much expected that the prescription can be generated automatically to assist doctors in prescribing using such as machine learning on the tongue images. However, it is confronted with two challenges. First, there is not a larger tongue image database available for machine learning. Second, there is no such machine learning method available for generating prescription according to the given tongue image. This paper begins with constructing a larger tongue image database, where each image corresponds to a prescription. It then uses auto-encoder to extract features for the tongue image, on which the recommendation neural network is proposed to recommend herbs for the prescription. Finally, a new prescription generation method is proposed to select optimal herbs from the recommended herbs to form the final prescription. Experimental results on our constructed databases validate the effectiveness and the superior performance of the proposed methods.

Machine learning algorithms in classifying TCM tongue features in diabetes mellitus and symptoms of gastric disease

IntroductionTongue diagnosis in Traditional Chinese Medicine (TCM) is used in clinical practice for disease diagnosis but this may be perceived as being too subjective. Objective technological approaches can lead to more accurate diagnosis and have become a focus for research. The aim of this study was to explore different tongue features in people with diabetes, gastric symptoms using images collected using digital tongue imaging. MethodsMachine learning algorithms were used to classify diabetes mellitus(DM) patients and healthy participants, and patients exhibiting two different TCM gastric disease symptoms (liver-stomach disharmony and spleen-stomach deficiency). Multi-type feature extraction and selection from 466 tongue images was conducted. In the feature extraction stage, texture features and four TCM tongue features were identified: constitution color, coating color, cracks, and plumpness and slenderness. In the classification stage, two different classification algorithms were employed, Random Forest and Support Vector Machine, to classify DM and TCM gastric disease symptoms. ResultsRegarding the classification results, the area under the receiver operating characteristic curve (AUC) reached 97.9% ± 0.1%, with maximum sensitivity and specificity of 93.5% ± 0.6% and 95.0% ± 0.8%, for DM patients and healthy volunteers respectively. For the classification of TCM gastric disease symptoms, the highest AUC was only 69.4% ± 2.6%, with a sensitivity and specificity of 65.2% ± 2.9% and 63.7% ± 2.1%, respectively. ConclusionsExperiments shows that the TCM tongue features (color, cracks, plumpness and slenderness) and texture features have good performance for disease diagnosis. However, classification of DM was significantly better than for TCM gastric symptoms. The RF classification model was also significantly better than the SVM classifier model.

Automatic Construction of Chinese Herbal Prescriptions From Tongue Images Using CNNs and Auxiliary Latent Therapy Topics

The tongue image provides important physical information of humans. It is of great importance for diagnoses and treatments in clinical medicine. Herbal prescriptions are simple, noninvasive, and have low side effects. Thus, they are widely applied in China. Studies on the automatic construction technology of herbal prescriptions based on tongue images have great significance for deep learning to explore the relevance of tongue images for herbal prescriptions, it can be applied to healthcare services in mobile medical systems. In order to adapt to the tongue image in a variety of photographic environments and construct herbal prescriptions, a neural network framework for prescription construction is designed. It includes single/double convolution channels and fully connected layers. Furthermore, it proposes the auxiliary therapy topic loss mechanism to model the therapy of Chinese doctors and alleviate the interference of sparse output labels on the diversity of results. The experiment use the real-world tongue images and the corresponding prescriptions and the results can generate prescriptions that are close to the real samples, which verifies the feasibility of the proposed method for the automatic construction of herbal prescriptions from tongue images. Also, it provides a reference for automatic herbal prescription construction from more physical information.

Diabetes Tongue Image Classification Using Machine Learning and Deep Learning

BackgroundIn China, diabetes is a common, high-incidence chronic disease. Diabetes has become a serious public health problem. However, the current diagnosis and treatment methods are difficult to control the progress of diabetes. Traditional Chinese Medicine (TCM) has become an option for the treatment of diabetes due to its low cost, satisfactory curative effect and good accessibility. Objective: Based on the tongue image data to realize the fine classification of the diabetic population, provide a diagnostic basis for the formulation of individualized treatment plans for diabetes, ensure the accuracy and consistency of the diagnosis of TCM, and promote the objective, standardized and standardized development of TCM diagnosis. Methods: We use the TFDA-1 tongue examination instrument to collect the tongue images of the subjects. Tongue Diagnosis Analysis System (TDAS) is used to extract the color feature, texture feature and tongue coating ratio feature of the tongue image. Auto-Encoder extracts auto-encoding features from tongue images through self-supervised learning extraction. We use K-means to perform fusion calculations on TDAS features and self-encoding features to classify tongue images. TDAS features are used to describe the differences between clusters and the characteristics of clusters. Vision Transformer (ViT) combined with Grad-weighted Class Activation Mapping (Grad-CAM) is used to verify the clustering results and calculate positioning diagnostic information. Results: According to the input tongue image data, K-means divides the diabetic population into 4 clusters with clear boundaries between clusters. Cluster 3 had the highest TB-L and TC-L, and Cluster 2 had the lowest TB-L and TC-L (P <0.001). Cluster 2 and Cluster 4 had the highest TB-a (P <0.01, P <0.001), and Cluster 3 had the lowest TB-a (P <0.001). Cluster 2 had the highest TB-b and TC-b (P <0.01, P <0.001), and Cluster 3 had the lowest TB-b and TC-b (P <0.01, P <0.001). Cluster 2 had the highest TB-ASM, and TB-CON, TB-ENT and TB-MEAN had the lowest (P <0.001). Cluster 3 had the lowest TC-ASM, TC-CON, TC-ENT and TC-MEAN the highest (P <0.01, P <0.001). Cluster 3 had the highest Per-all (P <0.001). Vision Transformer (ViT) verifies the clustering results of K-means, the highest Top-1 Classification Accuracy (CA) is 84.9%, and the average CA is 83.6%. Conclusions: The study organically combined unsupervised learning, self-supervised learning and supervised learning, and designed a complete diabetic tongue image classification method. This method does not rely on human intervention, and makes decisions based entirely on tongue image data and achieves state-of-the-art results. Our research will help TCM deeply participate in the individualized treatment of diabetes, and provide new ideas for promoting the objective, standardization and standardization of TCM diagnosis. Funding: The authors gratefully acknowledge the financial supports by the National Key Research and Development Program of China [grant numbers 2017YFC1703300 and 2017YFC1703301]; National Natural Science Foundation of China [grant numbers 81873235, 81973750, 82004258 and 81904094]; and 1226 Engineering Health Key Project [grant number BWS17J028]. Declaration of Interest: We declare no competing interests. Ethical Approval: The study was approved by the IRB of Shuguang Hospital affiliated with Shanghai University of TCM.

Internet of Things and Synergic Deep Learning Based Biomedical Tongue Color Image Analysis for Disease Diagnosis and Classification

In recent times, internet of things (IoT) and wireless communication techniques become widely used in healthcare sector. Biomedical image processing is commonly employed to detect the existence of diseases using biomedical images. Tongue diagnosis is an efficient, non-invasive model to perform auxiliary diagnosis any time anywhere that is support the global necessity in the primary healthcare system. Conventionally, medical practitioners investigate the tongue features based on their expert's knowledge comes from experience. In order to eradicate the qualitative aspects, tongue images can be quantitatively examined, offering an effective disease diagnostic process in such a way that the physical harm of the patients can be minimized. Numerous tongue image analysis approaches exist in the literature, it is required to develop automated deep learning (DL) models to diagnose the diseases using tongue image analysis. In this view, this paper designs an automated IoT and synergic deep learning based tongue color image (ASDL-TCI) analysis model for disease diagnosis and classification. The proposed ASDL-TCI model operates on major stages namely data acquisition, pre-processing, feature extraction, classification, and parameter optimization. Primarily, the IoT devices are used to capture the human tongue images and transmitted to the cloud for further analysis. In addition, median filtering based image pre-processing and SDL based feature extraction techniques are employed. Moreover, deep neural network (DNN) based classifier is applied to determine the existence of the diseases. Lastly, enhanced black widow optimization (EBWO) based parameter tuning process takes place to enhance the diagnostic performance. For assessing the effectual performance of the ASDL-TCI model, a set of simulations take place on benchmark tongue images and examined the results under distinct dimensions. The simulation outcome verified the enhanced diagnostic performance of the ASDL-TCI model over the compared methods with the maximum precision, recall, and accuracy of 0.984, 0.973, and 0.983.

LSM-SEC: Tongue Segmentation by the Level Set Model with Symmetry and Edge Constraints.

Accurate segmentation of the tongue body is an important prerequisite for computer-aided tongue diagnosis. In general, the size and shape of the tongue are very different, the color of the tongue is similar to the surrounding tissue, the edge of the tongue is fuzzy, and some of the tongue is interfered by pathological details. The existing segmentation methods are often not ideal for tongue image processing. To solve these problems, this paper proposes a symmetry and edge-constrained level set model combined with the geometric features of the tongue for tongue segmentation. Based on the symmetry geometry of the tongue, a novel level set initialization method is proposed to improve the accuracy of subsequent model evolution. In order to increase the evolution force of the energy function, symmetry detection constraints are added to the evolution model. Combined with the latest convolution neural network, the edge probability input of the tongue image is obtained to guide the evolution of the edge stop function, so as to achieve accurate and automatic tongue segmentation. The experimental results show that the input tongue image is not subject to the external capturing facility or environment, and it is suitable for tongue segmentation under most realistic conditions. Qualitative and quantitative comparisons show that the proposed method is superior to the other methods in terms of robustness and accuracy.

Improving ultrasound-based multimodal speech recognition with predictive features from representation learning.

Representation learning is believed to produce high-level representations of underlying dynamics in temporal sequences. A three-dimensional convolutional neural network trained to predict future frames in ultrasound tongue and optical lip images creates features for a continuous hidden Markov model based speech recognition system. Predictive tongue features are found to generate lower word error rates than those obtained from an auto-encoder without future frames, or from discrete cosine transforms. Improvement is apparent for the monophone/triphone Gaussian mixture model and deep neural network acoustic models. When tongue and lip modalities are combined, the advantage of the predictive features is reduced.

Creating Song From Lip and Tongue Videos With a Convolutional Vocoder

A convolutional neural network and deep autoencoder are used to predict Line Spectral Frequencies, F0, and a voiced/unvoiced flag in singing data, using as input only ultrasound images of the tongue and visual images of the lips. A novel convolutional vocoder to transform the learned parameters into an audio signal is also presented. Spectral Distortion of predicted Line Spectral Frequencies is reduced compared to that in an earlier study using handcrafted features and multilayer perceptrons on the same data set; while predicted F0 and voiced/unvoiced flag predictions are found to be highly correlated with their ground truth values. Comparison of the convolutional vocoder to standard vocoders is made. Results can be of interest in the study of singing articulation as well as for silent speech interface research. Sample predicted audio files are available online. Source code: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/TjuJianyu/SSI_DL</uri> .

Analysis of Tongue Color-Associated Features among Patients with PCR-Confirmed COVID-19 Infection in Ukraine

Objective: To evaluate and systematize tongue color-related manifestations among patients with PCR-confirmed COVID-19 infection. Material and Methods: This retrospective study included analysis of tongue images obtained from patients with PCR-confirmed COVID-19 infection. Evaluation of coronavirus disease severity (mild, moderate, severe, critical) was provided, considering clinical symptomatology and results of laboratorial and instrumental diagnostic methods. Each picture was analyzed considering the parameters of color of the tongue and color of the tongue plaque by two dental specialists. Cochran-Armitage test for trend was used to evaluate associations between the tongue color and tongue plaque color, and coronavirus disease severity. Results: The most prevalent tongue colors were pale pink, red and dark red (burgundy color). A total of 64.29% of patients with mild disease demonstrated pale pink color of the tongue. Patients with moderate coronavirus disease were characterized with the adverse trend: 62.35% of them presented with red-colored tongue, while in 37.64% of cases, the tongue was pale pink. Severe COVID-19 patients, almost in 90% of the cases, had either red or burgundy color of the tongue. Conclusion: SARS-COV-2 infection is not manifested by tongue-targeted or tongue-specific signs and features; however, coronavirus disease itself provokes changes within the tongue color and tongue plaque color similar to those registered during other internal pathologies.

Deep sparse transfer learning for remote smart tongue diagnosis.

People are exploring new ideas based on artificial intelligent infrastructures for immediate processing, in which the main obstacles of widely-deploying deep methods are the huge volume of neural network and the lack of training data. To meet the high computing and low latency requirements in modeling remote smart tongue diagnosis with edge computing, an efficient and compact deep neural network design is necessary, while overcoming the vast challenge on modeling its intrinsic diagnosis patterns with the lack of clinical data. To address this challenge, a deep transfer learning model is proposed for the effective tongue diagnosis, based on the proposed similar-sparse domain adaptation (SSDA) scheme. Concretely, a transfer strategy of similar data is introduced to efficiently transfer necessary knowledge, overcoming the insufficiency of clinical tongue images. Then, to generate simplified structure, the network is pruned with transferability remained in domain adaptation. Finally, a compact model combined with two sparse networks is designed to match limited edge device. Extensive experiments are conducted on the real clinical dataset. The proposed model can use fewer training data samples and parameters to produce competitive results with less power and memory consumptions, making it possible to widely run smart tongue diagnosis on low-performance infrastructures.