International Joint Conference on Neural Networks

The standard chronic wound assessment method based on visual examination is potentially inaccurate and also represents a significant clinical workload. Hence, computer-based systems providing quantitative wound assessment may be valuable for accurately monitoring wound healing status, with the wound area the best suited for automated analysis. Here, we present a novel approach, using support vector machines (SVM) to determine the wound boundaries on foot ulcer images captured with an image capture box, which provides controlled lighting and range. After superpixel segmentation, a cascaded two-stage classifier operates as follows: in the first stage, a set of k binary SVM classifiers are trained and applied to different subsets of the entire training images dataset, and incorrectly classified instances are collected. In the second stage, another binary SVM classifier is trained on the incorrectly classified set. We extracted various color and texture descriptors from superpixels that are used as input for each stage in the classifier training. Specifically, color and bag-of-word representations of local dense scale invariant feature transformation features are descriptors for ruling out irrelevant regions, and color and wavelet-based features are descriptors for distinguishing healthy tissue from wound regions. Finally, the detected wound boundary is refined by applying the conditional random field method. We have implemented the wound classification on a Nexus 5 smartphone platform, except for training which was done offline. Results are compared with other classifiers and show that our approach provides high global performance rates (average sensitivity = 73.3%, specificity = 94.6%) and is sufficiently efficient for a smartphone-based image analysis.

Area Determination of Diabetic Foot Ulcer Images Using a Cascaded Two-Stage SVM-Based Classification

Acute and chronic wounds have varying etiologies and are an economic burden to healthcare systems around the world. The advanced wound care market is expected to exceed $22 billion by 2024. Wound care professionals rely heavily on images and image documentation for proper diagnosis and treatment. Unfortunately lack of expertise can lead to improper diagnosis of wound etiology and inaccurate wound management and documentation. Fully automatic segmentation of wound areas in natural images is an important part of the diagnosis and care protocol since it is crucial to measure the area of the wound and provide quantitative parameters in the treatment. Various deep learning models have gained success in image analysis including semantic segmentation. This manuscript proposes a novel convolutional framework based on MobileNetV2 and connected component labelling to segment wound regions from natural images. The advantage of this model is its lightweight and less compute-intensive architecture. The performance is not compromised and is comparable to deeper neural networks. We build an annotated wound image dataset consisting of 1109 foot ulcer images from 889 patients to train and test the deep learning models. We demonstrate the effectiveness and mobility of our method by conducting comprehensive experiments and analyses on various segmentation neural networks. The full implementation is available at https://github.com/uwm-bigdata/wound-segmentation .

/pdf/fully-automatic-wound-segmentation-with-deep-convolutional-2blmomnaw6.pdf

Fully automatic wound segmentation with deep convolutional neural networks

Computerized segmentation and measurement of chronic wound images

BACKGROUND: Each year more than 10 million people worldwide are burned severely enough to require medical attention, with clinical outcomes noticeably worse in resource poor settings. Expert clinical advice on acute injuries can play a determinant role and there is a need for novel approaches that allow for timely access to advice. We developed an interactive mobile phone application that enables transfer of both patient data and pictures of a wound from the point-of-care to a remote burns expert who, in turn, provides advice back. METHODS AND RESULTS: The application is an integrated clinical decision support system that includes a mobile phone application and server software running in a cloud environment. The client application is installed on a smartphone and structured patient data and photographs can be captured in a protocol driven manner. The user can indicate the specific injured body surface(s) through a touchscreen interface and an integrated calculator estimates the total body surface area that the burn injury affects. Predefined standardised care advice including total fluid requirement is provided immediately by the software and the case data are relayed to a cloud server. A text message is automatically sent to a burn expert on call who then can access the cloud server with the smartphone app or a web browser, review the case and pictures, and respond with both structured and personalized advice to the health care professional at the point-of-care. CONCLUSIONS: In this article, we present the design of the smartphone and the server application alongside the type of structured patient data collected together with the pictures taken at point-of-care. We report on how the application will be introduced at point-of-care and how its clinical impact will be evaluated prior to roll out. Challenges, strengths and limitations of the system are identified that may help materialising or hinder the expected outcome to provide a solution for remote consultation on burns that can be integrated into routine acute clinical care and thereby promote equity in injury emergency care, a growing public health burden. Language: en

/pdf/a-smartphone-app-and-cloud-based-consultation-system-for-2av7nm1jy7.pdf

A Smartphone App and Cloud-Based Consultation System for Burn Injury Emergency Care

When burn injury occurs, the most important step is to provide treatment to the injury immediately by identifying degree of the burn which can only be diagnosed by specialists. However, specialists for burn trauma are still inadequate for some local hospitals. Hence, the invention of an automatic system that is able to help evaluating the burn would be extremely beneficial to those hospitals. The aim of this work is to develop an automatic system with the ability of providing the first assessment to burn injury from burn color images. The method used in this work can be divided into 2 parts, i.e., burn image segmentation and degree of burn identification. Burn image segmentation employs the Cr-transformation, Luv-transformation and fuzzy c-means clustering technique to separate the burn wound area from healthy skin and then mathematical morphology is applied to reduce segmentation errors. The segmentation algorithm performance is evaluated by the positive predictive value (PPV) and the sensitivity (S). Burn degree identification uses h-transformation and texture analysis to extract feature vectors and the support vector machine (SVM) is applied to identify the degree of burn. The classification results are compared with that of Bayes and K-nearest neighbor classifiers. The experimental results show that our proposed segmentation algorithm yields good results for the burn color images. The PPV and S are about 0.92 and 0.84, respectively. Degree of burn identification experiments show that SVM yields the best results of 89.29 % correct classification on the validation sets of the 4-fold cross validation. SVM also yields 75.33 % correct classification on the blind test experiment.

Automatic segmentation and degree identification in burn color images

This paper presents a river flood prediction technique using support vector machine (SVM). We investigated the 2-year data covering 2005 and 2006 and 7 crucial river floods that occurred in the downtown of Chiang Mai, Thailand. Past and current river levels of the 3 gauging stations are utilized as the input data of the SVM models to predict the river levels at the downtown station in 1 hour and 7 hours in advance. The performances of the SVM models are compared with that of the multilayer perceptrons (MLP) models. The experimental results show that the SVM models can perform better than the MLP models. Moreover, the results from the blind test sets demonstrate that the SVM models are appropriate for warning people before flood events. The proposed SVM prediction models are also implemented in a real-world flood warning system. The predicted river levels are accessible to public via a Web site, electronic billboards, and warning stations all over the city.

River basin flood prediction using support vector machines

Automatic cardiac T2* relaxation time estimation from magnetic resonance images using region growing method with automatically initialized seed points

Dental age estimation is considered to be an accurate method for identifying the age of an unknown person. This study aims to apply a deep-learning algorithm to assess mandibular third molar develo...

Semi-automated technique to assess the developmental stage of mandibular third molars for age estimation

Problem statement: Estimation of moisture contents of dried food products from their
dielectric constants was an important step in moisture measurement systems. The regression models
that provide good prediction performance are desirable. Approach: The Multilayer Perceptrons (MLP)
and Support Vector Regression (SVR) were applied in this research to predict the moisture contents of
dried longan arils from their dielectric constants. The data set was collected from 1500 samples of
dried longan aril with five different moisture contents of 10, 14, 18, 22 and 25% Wet basis (Wb.)
Dielectric constant of dried longan aril was measured by using our previously proposed electrical
capacitance-based system. The results from the MLP and SVR models were compared to that from the
linear regression and polynomial regression models. To take into account the generalization of the
models, the four-fold cross validation was applied. Results: For the training sets, the average mean
absolute errors over three bulk densities of 1.30, 1.45 and 1.60 g cm-3 were 1.7578, 0.6157, 0.3812,
0.3113, 0.0103 and 0.0044% Wb for the linear regression, second-, third-, fourth-order polynomial
regression, MLP and SVR models, respectively. For the validation sets, the average mean absolute
errors over the three bulk densities were 1.7616, 0.6192, 0.3844, 0.3146, 0.0126 and 0.0093% Wb for
the linear regression, 2nd, 3rd and 4th-order polynomial regression, MLP and SVR models,
respectively. Conclusion: The regression models based on MLP and SVR yielded better performances
than the models based on linear regression and polynomial regression on both training and validation
sets. The models based on MLP and SVR also provided robustness to the variation of bulk density. Not
only for dried longan aril, the proposed models can also be adapted and applied to other materials or
dried food products.

/pdf/moisture-content-prediction-of-dried-longan-aril-from-3csf2feq82.pdf

Kittichai Wantanajittikul

Papers

Automatic segmentation and degree identification in burn color images

River basin flood prediction using support vector machines

Automatic cardiac T2* relaxation time estimation from magnetic resonance images using region growing method with automatically initialized seed points

Semi-automated technique to assess the developmental stage of mandibular third molars for age estimation

Moisture Content Prediction of Dried Longan Aril from Dielectric Constant Using Multilayer Perceptrons and Support Vector Regression