Toward Clinically Applicable 3-Dimensional Tooth Segmentation via Deep Learning

Digital dentistry plays a pivotal role in dental health care. A critical step in many digital dental systems is to accurately delineate individual teeth and the gingiva in the 3-dimension intraoral scanned mesh data. However, previous state-of-the-art methods are either time-consuming or error prone, hence hindering their clinical applicability. This article presents an accurate, efficient, and fully automated deep learning model trained on a data set of 4,000 intraoral scanned data annotated by experienced human experts. On a holdout data set of 200 scans, our model achieves a per-face accuracy, average-area accuracy, and area under the receiver operating characteristic curve of 96.94%, 98.26%, and 0.9991, respectively, significantly outperforming the state-of-the-art baselines. In addition, our model takes only about 24 s to generate segmentation outputs, as opposed to >5 min by the baseline and 15 min by human experts. A clinical performance test of 500 patients with malocclusion and/or abnormal teeth shows that 96.9% of the segmentations are satisfactory for clinical applications, 2.9% automatically trigger alarms for human improvement, and only 0.2% of them need rework. Our research demonstrates the potential for deep learning to improve the efficacy and efficiency of dental treatment and digital dentistry.

Download Full-text

Clinically Applicable System For 3D Teeth Segmentation in Intraoral Scans using Deep Learning

10.21203/rs.3.rs-103285/v1 ◽

2020 ◽

Author(s):

Jin Hao ◽

Wen Liao ◽

Yueling Zhang ◽

Peilin Li ◽

Jianru Yi ◽

...

Keyword(s):

Deep Learning ◽

Performance Test ◽

Dental Treatment ◽

State Of The Art ◽

Clinical Performance ◽

Characteristic Curve ◽

Three Dimension ◽

Digital Dentistry ◽

Previous State ◽

Deep Learning Model

Abstract Digital dentistry plays a pivotal role in dental healthcare. A critical step in many digital dental systems is to accurately delineate individual teeth and the gingiva in the three-dimension (3D) intraoral scanned (IOS) mesh data. However, previous state-of-the-art methods are either time-consuming or error-prone, hence hinder their clinical applicability. In this paper, we present an accurate, efficient, and fully-automated deep learning model, trained on a dataset of 4,000 IOS data annotated by experienced human experts. On a hold-out dataset of 200 scans, our model achieves a per-face accuracy, average-area accuracy and area under the receiver operating characteristic curve (AUC) of 96.94%, 98.26%, and 0.9991, respectively, significantly outperforming the state-of-the-art baseline. In addition, our model only takes about 24 seconds to generate segmentation outputs, as compared to over 5 minutes by the baseline and 15 minutes by human experts. A clinical performance test of 500 patients with malocclusion or/and abnormal teeth shows that 96.9% of the segmentations are satisfactory for clinical applications, 2.9% automatically trigger alarms for human improvement, and only 0.2% of them need rework. Our research demonstrates the potential for deep learning to improve the efficacy and efficiency of dental treatment and digital dentistry.

Download Full-text

A Study on 3D Deep Learning-Based Automatic Diagnosis of Nasal Fractures

Sensors ◽

10.3390/s22020506 ◽

2022 ◽

Vol 22 (2) ◽

pp. 506

Author(s):

Yu-Jin Seol ◽

Young-Jae Kim ◽

Yoon-Sang Kim ◽

Young-Woo Cheon ◽

Kwang-Gi Kim

Keyword(s):

Deep Learning ◽

Binary Classification ◽

Characteristic Curve ◽

Nasal Bone ◽

Facial Trauma ◽

Future Research ◽

Automatic Diagnosis ◽

3 Dimensional ◽

Computed Tomography Images ◽

Nasal Fractures

This paper reported a study on the 3-dimensional deep-learning-based automatic diagnosis of nasal fractures. (1) Background: The nasal bone is the most protuberant feature of the face; therefore, it is highly vulnerable to facial trauma and its fractures are known as the most common facial fractures worldwide. In addition, its adhesion causes rapid deformation, so a clear diagnosis is needed early after fracture onset. (2) Methods: The collected computed tomography images were reconstructed to isotropic voxel data including the whole region of the nasal bone, which are represented in a fixed cubic volume. The configured 3-dimensional input data were then automatically classified by the deep learning of residual neural networks (3D-ResNet34 and ResNet50) with the spatial context information using a single network, whose performance was evaluated by 5-fold cross-validation. (3) Results: The classification of nasal fractures with simple 3D-ResNet34 and ResNet50 networks achieved areas under the receiver operating characteristic curve of 94.5% and 93.4% for binary classification, respectively, both indicating unprecedented high performance in the task. (4) Conclusions: In this paper, it is presented the possibility of automatic nasal bone fracture diagnosis using a 3-dimensional Resnet-based single classification network and it will improve the diagnostic environment with future research.

Download Full-text

Classification of Clinically Significant Prostate Cancer on Multi-Parametric MRI: A Validation Study Comparing Deep Learning and Radiomics

Cancers ◽

10.3390/cancers14010012 ◽

2021 ◽

Vol 14 (1) ◽

pp. 12

Author(s):

Jose M. Castillo T. ◽

Muhammad Arif ◽

Martijn P. A. Starmans ◽

Wiro J. Niessen ◽

Chris H. Bangma ◽

...

Keyword(s):

Prostate Cancer ◽

Deep Learning ◽

Characteristic Curve ◽

Model Development ◽

Learning Model ◽

Multiparametric Mri ◽

Data Sets ◽

Data Set ◽

Test Sets ◽

Deep Learning Model

The computer-aided analysis of prostate multiparametric MRI (mpMRI) could improve significant-prostate-cancer (PCa) detection. Various deep-learning- and radiomics-based methods for significant-PCa segmentation or classification have been reported in the literature. To be able to assess the generalizability of the performance of these methods, using various external data sets is crucial. While both deep-learning and radiomics approaches have been compared based on the same data set of one center, the comparison of the performances of both approaches on various data sets from different centers and different scanners is lacking. The goal of this study was to compare the performance of a deep-learning model with the performance of a radiomics model for the significant-PCa diagnosis of the cohorts of various patients. We included the data from two consecutive patient cohorts from our own center (n = 371 patients), and two external sets of which one was a publicly available patient cohort (n = 195 patients) and the other contained data from patients from two hospitals (n = 79 patients). Using multiparametric MRI (mpMRI), the radiologist tumor delineations and pathology reports were collected for all patients. During training, one of our patient cohorts (n = 271 patients) was used for both the deep-learning- and radiomics-model development, and the three remaining cohorts (n = 374 patients) were kept as unseen test sets. The performances of the models were assessed in terms of their area under the receiver-operating-characteristic curve (AUC). Whereas the internal cross-validation showed a higher AUC for the deep-learning approach, the radiomics model obtained AUCs of 0.88, 0.91 and 0.65 on the independent test sets compared to AUCs of 0.70, 0.73 and 0.44 for the deep-learning model. Our radiomics model that was based on delineated regions resulted in a more accurate tool for significant-PCa classification in the three unseen test sets when compared to a fully automated deep-learning model.

Download Full-text

Ensemble-based deep meta learning for medical image segmentation

Journal of Intelligent & Fuzzy Systems ◽

10.3233/jifs-219221 ◽

2021 ◽

pp. 1-7

Author(s):

Usman Ahmed ◽

Jerry Chun-Wei Lin ◽

Gautam Srivastava

Keyword(s):

Image Segmentation ◽

Deep Learning ◽

Learning Algorithm ◽

State Of The Art ◽

Medical Image Segmentation ◽

The State ◽

Data Set ◽

Deep Learning Algorithm ◽

Meta Learning ◽

Small Set

Deep learning methods have led to a state of the art medical applications, such as image classification and segmentation. The data-driven deep learning application can help stakeholders to collaborate. However, limited labelled data set limits the deep learning algorithm to generalize for one domain into another. To handle the problem, meta-learning helps to learn from a small set of data. We proposed a meta learning-based image segmentation model that combines the learning of the state-of-the-art model and then used it to achieve domain adoption and high accuracy. Also, we proposed a prepossessing algorithm to increase the usability of the segments part and remove noise from the new test image. The proposed model can achieve 0.94 precision and 0.92 recall. The ability to increase 3.3% among the state-of-the-art algorithms.

Download Full-text

Multi-Hazard Exposure Mapping Using Machine Learning Techniques: A Case Study from Iran

Remote Sensing ◽

10.3390/rs11161943 ◽

2019 ◽

Vol 11 (16) ◽

pp. 1943 ◽

Cited By ~ 15

Author(s):

Omid Rahmati ◽

Saleh Yousefi ◽

Zahra Kalantari ◽

Evelyn Uuemaa ◽

Teimur Teimurian ◽

...

Keyword(s):

Machine Learning ◽

State Of The Art ◽

Characteristic Curve ◽

Machine Learning Techniques ◽

Support Vector ◽

Mountainous Area ◽

Data Set ◽

Boosted Regression Tree ◽

Hazard Exposure ◽

Learning Techniques

Mountainous areas are highly prone to a variety of nature-triggered disasters, which often cause disabling harm, death, destruction, and damage. In this work, an attempt was made to develop an accurate multi-hazard exposure map for a mountainous area (Asara watershed, Iran), based on state-of-the art machine learning techniques. Hazard modeling for avalanches, rockfalls, and floods was performed using three state-of-the-art models—support vector machine (SVM), boosted regression tree (BRT), and generalized additive model (GAM). Topo-hydrological and geo-environmental factors were used as predictors in the models. A flood dataset (n = 133 flood events) was applied, which had been prepared using Sentinel-1-based processing and ground-based information. In addition, snow avalanche (n = 58) and rockfall (n = 101) data sets were used. The data set of each hazard type was randomly divided to two groups: Training (70%) and validation (30%). Model performance was evaluated by the true skill score (TSS) and the area under receiver operating characteristic curve (AUC) criteria. Using an exposure map, the multi-hazard map was converted into a multi-hazard exposure map. According to both validation methods, the SVM model showed the highest accuracy for avalanches (AUC = 92.4%, TSS = 0.72) and rockfalls (AUC = 93.7%, TSS = 0.81), while BRT demonstrated the best performance for flood hazards (AUC = 94.2%, TSS = 0.80). Overall, multi-hazard exposure modeling revealed that valleys and areas close to the Chalous Road, one of the most important roads in Iran, were associated with high and very high levels of risk. The proposed multi-hazard exposure framework can be helpful in supporting decision making on mountain social-ecological systems facing multiple hazards.

Download Full-text

Diagnostic assessment of a deep learning system for detecting atrial fibrillation in pulse waveforms

Heart ◽

10.1136/heartjnl-2018-313147 ◽

2018 ◽

Vol 104 (23) ◽

pp. 1921-1928 ◽

Cited By ~ 36

Author(s):

Ming-Zher Poh ◽

Yukkee Cheung Poh ◽

Pak-Hei Chan ◽

Chun-Ka Wong ◽

Louise Pun ◽

...

Keyword(s):

Atrial Fibrillation ◽

Deep Learning ◽

Test Data ◽

Predictive Value ◽

Characteristic Curve ◽

Performance Comparison ◽

Learning System ◽

Training Data ◽

Validation Data ◽

Data Set

ObjectiveTo evaluate the diagnostic performance of a deep learning system for automated detection of atrial fibrillation (AF) in photoplethysmographic (PPG) pulse waveforms.MethodsWe trained a deep convolutional neural network (DCNN) to detect AF in 17 s PPG waveforms using a training data set of 149 048 PPG waveforms constructed from several publicly available PPG databases. The DCNN was validated using an independent test data set of 3039 smartphone-acquired PPG waveforms from adults at high risk of AF at a general outpatient clinic against ECG tracings reviewed by two cardiologists. Six established AF detectors based on handcrafted features were evaluated on the same test data set for performance comparison.ResultsIn the validation data set (3039 PPG waveforms) consisting of three sequential PPG waveforms from 1013 participants (mean (SD) age, 68.4 (12.2) years; 46.8% men), the prevalence of AF was 2.8%. The area under the receiver operating characteristic curve (AUC) of the DCNN for AF detection was 0.997 (95% CI 0.996 to 0.999) and was significantly higher than all the other AF detectors (AUC range: 0.924–0.985). The sensitivity of the DCNN was 95.2% (95% CI 88.3% to 98.7%), specificity was 99.0% (95% CI 98.6% to 99.3%), positive predictive value (PPV) was 72.7% (95% CI 65.1% to 79.3%) and negative predictive value (NPV) was 99.9% (95% CI 99.7% to 100%) using a single 17 s PPG waveform. Using the three sequential PPG waveforms in combination (<1 min in total), the sensitivity was 100.0% (95% CI 87.7% to 100%), specificity was 99.6% (95% CI 99.0% to 99.9%), PPV was 87.5% (95% CI 72.5% to 94.9%) and NPV was 100% (95% CI 99.4% to 100%).ConclusionsIn this evaluation of PPG waveforms from adults screened for AF in a real-world primary care setting, the DCNN had high sensitivity, specificity, PPV and NPV for detecting AF, outperforming other state-of-the-art methods based on handcrafted features.

Download Full-text

Unsupervised Representation High-Resolution Remote Sensing Image Scene Classification via Contrastive Learning Convolutional Neural Network

Photogrammetric Engineering & Remote Sensing ◽

10.14358/pers.87.8.577 ◽

2021 ◽

Vol 87 (8) ◽

pp. 577-591

Author(s):

Fengpeng Li ◽

Jiabao Li ◽

Wei Han ◽

Ruyi Feng ◽

Lizhe Wang

Keyword(s):

Neural Network ◽

Remote Sensing ◽

Deep Learning ◽

High Resolution ◽

Convolutional Neural Network ◽

State Of The Art ◽

Remote Sensing Image ◽

Scene Classification ◽

Data Set ◽

Unsupervised Deep Learning

Inspired by the outstanding achievement of deep learning, supervised deep learning representation methods for high-spatial-resolution remote sensing image scene classification obtained state-of-the-art performance. However, supervised deep learning representation methods need a considerable amount of labeled data to capture class-specific features, limiting the application of deep learning-based methods while there are a few labeled training samples. An unsupervised deep learning representation, high-resolution remote sensing image scene classification method is proposed in this work to address this issue. The proposed method, called contrastive learning, narrows the distance between positive views: color channels belonging to the same images widens the gaps between negative view pairs consisting of color channels from different images to obtain class-specific data representations of the input data without any supervised information. The classifier uses extracted features by the convolutional neural network (CNN)-based feature extractor with labeled information of training data to set space of each category and then, using linear regression, makes predictions in the testing procedure. Comparing with existing unsupervised deep learning representation high-resolution remote sensing image scene classification methods, contrastive learning CNN achieves state-of-the-art performance on three different scale benchmark data sets: small scale RSSCN7 data set, midscale aerial image data set, and large-scale NWPU-RESISC45 data set.

Download Full-text

An annotation-free whole-slide training approach to pathological classification of lung cancer types using deep learning

Nature Communications ◽

10.1038/s41467-021-21467-y ◽

2021 ◽

Vol 12 (1) ◽

Author(s):

Chi-Long Chen ◽

Chi-Chung Chen ◽

Wei-Hsiang Yu ◽

Szu-Hua Chen ◽

Yu-Chan Chang ◽

...

Keyword(s):

Lung Cancer ◽

Deep Learning ◽

Characteristic Curve ◽

Digital Pathology ◽

Classification Performance ◽

Multiple Instance Learning ◽

Data Set ◽

Memory Mechanism ◽

Strong Localization ◽

Training Approach

AbstractDeep learning for digital pathology is hindered by the extremely high spatial resolution of whole-slide images (WSIs). Most studies have employed patch-based methods, which often require detailed annotation of image patches. This typically involves laborious free-hand contouring on WSIs. To alleviate the burden of such contouring and obtain benefits from scaling up training with numerous WSIs, we develop a method for training neural networks on entire WSIs using only slide-level diagnoses. Our method leverages the unified memory mechanism to overcome the memory constraint of compute accelerators. Experiments conducted on a data set of 9662 lung cancer WSIs reveal that the proposed method achieves areas under the receiver operating characteristic curve of 0.9594 and 0.9414 for adenocarcinoma and squamous cell carcinoma classification on the testing set, respectively. Furthermore, the method demonstrates higher classification performance than multiple-instance learning as well as strong localization results for small lesions through class activation mapping.

Download Full-text

An interpretable framework for investigating the neighborhood effect in POI recommendation

PLoS ONE ◽

10.1371/journal.pone.0255685 ◽

2021 ◽

Vol 16 (8) ◽

pp. e0255685

Author(s):

Guangchao Yuan ◽

Munindar P. Singh ◽

Pradeep K. Murukannaiah

Keyword(s):

Deep Learning ◽

Matrix Factorization ◽

State Of The Art ◽

Experimental Results ◽

Neighborhood Effect ◽

Data Set ◽

Point Of Interest ◽

Poi Recommendation ◽

Geographical Characteristics

Geographical characteristics have been proven to be effective in improving the quality of point-of-interest (POI) recommendation. However, existing works on POI recommendation focus on cost (time or money) of travel for a user. An important geographical aspect that has not been studied adequately is the neighborhood effect, which captures a user’s POI visiting behavior based on the user’s preference not only to a POI, but also to the POI’s neighborhood. To provide an interpretable framework to fully study the neighborhood effect, first, we develop different sets of insightful features, representing different aspects of neighborhood effect. We employ a Yelp data set to evaluate how different aspects of the neighborhood effect affect a user’s POI visiting behavior. Second, we propose a deep learning–based recommendation framework that exploits the neighborhood effect. Experimental results show that our approach is more effective than two state-of-the-art matrix factorization–based POI recommendation techniques.

Download Full-text

Detection of acne by deep learning object detection

10.1101/2021.12.05.21267310 ◽

2021 ◽

Author(s):

Amandip Sangha ◽

Mohammad Rizvi

Keyword(s):

Deep Learning ◽

Object Detection ◽

State Of The Art ◽

Average Precision ◽

Single Class ◽

Data Set ◽

Detection Model ◽

Model Predictions ◽

Deep Learning Model ◽

Facial Acne

AbstractImportanceState-of-the art performance is achieved with a deep learning object detection model for acne detection. There is little current research on object detection in dermatology and acne in particular. As such, this work is early in this field and achieves state of the art performance.ObjectiveTrain an object detection model on a publicly available data set of acne photos.Design, Setting, and ParticipantsA deep learning model is trained with cross validation on a data set of facial acne photos.Main Outcomes and MeasuresObject detection models for detecting acne for single-class (acne) and multi-class (four severity levels). We train and evaluate the models using standard metrics such as mean average precision (mAP). Then we manually evaluate the model predictions on the test set, and calculate accuracy in terms of precision, recall, F1, true and false positive and negative detections.ResultsWe achieve state-of-the art mean average precision [email protected] value of 37.97 for the single class acne detection task, and 26.50 for the 4-class acne detection task. Moreover, our manual evaluation shows that the single class detection model performs well on the validation set, achieving true positive 93.59 %, precision 96.45 % and recall 94.73 %.Conclusions and RelevanceWe are able to train a high-accuracy acne detection model using only a small publicly available data set of facial acne. Transfer learning on the pre-trained deep learning model yields good accuracy and high degree of transferability to patient submitted photographs. We also note that the training of standard architecture object detection models has given significantly better accuracy than more intricate and bespoke neural network architectures in the existing research literature.Key PointsQuestionCan deep learning-based acne detection models trained on a small data set of publicly available photos of patients with acne achieve high prediction accuracy?FindingsWe find that it is possible to train a reasonably good object detection model on a small, annotated data set of acne photos using standard deep learning architectures.MeaningDeep learning-based object detection models for acne detection can be a useful decision support tools for dermatologists treating acne patients in a digital clinical practice. It can prove a particularly useful tool for monitoring the time evolution of the acne disease state over prolonged time during follow-ups, as the model predictions give a quantifiable and comparable output for photographs over time. This is particularly helpful in teledermatological consultations, as a prediction model can be integrated in the patient-doctor remote communication.

Download Full-text