Plantar Pressure Images Research Articles

Deep-segmentation of plantar pressure images incorporating fully convolutional neural networks

Comfort shoe-last design relies on the key points of last curvature. Traditional plantar pressure image segmentation methods are limited to their local and global minimization issues. In this work, an improved fully convolutional networks (FCN) employing SegNet (SegNet+FCN 8 s) is proposed. The algorithm design and operation are performed using the visual geometry group (VGG). The method has high efficiency for the segmentation in positive indices of global accuracy (0.8105), average accuracy (0.8015), and negative indices of average cross-ratio (0.6110) and boundary F1 index (0.6200). The research has potential applications in improving the comfort of shoes.

Learning Models for Semantic Classification of Insufficient Plantar Pressure Images

Establishing a reliable and stable model to predict a target by using insufficient labeled samples is feasible and effective, particularly, for a sensor-generated data-set. This paper has been inspired with insufficient data-set learning algorithms, such as metric-based, prototype networks and meta-learning, and therefore we propose an insufficient data-set transfer model learning method. Firstly, two basic models for transfer learning are introduced. A classification system and calculation criteria are then subsequently introduced. Secondly, a dataset of plantar pressure for comfort shoe design is acquired and preprocessed through foot scan system; and by using a pre-trained convolution neural network employing AlexNet and convolution neural network (CNN)- based transfer modeling, the classification accuracy of the plantar pressure images is over 93.5%. Finally, the proposed method has been compared to the current classifiers VGG, ResNet, AlexNet and pre-trained CNN. Also, our work is compared with known-scaling and shifting (SS) and unknown-plain slot (PS) partition methods on the public test databases: SUN, CUB, AWA1, AWA2, and aPY with indices of precision (tr, ts, H) and time (training and evaluation). The proposed method for the plantar pressure classification task shows high performance in most indices when comparing with other methods. The transfer learning-based method can be applied to other insufficient data-sets of sensor imaging fields.

Plantar pressure image fusion for comfort fusion in diabetes mellitus using an improved fuzzy hidden Markov model

Diabetes mellitus is a clinical syndrome caused by the interaction of genetic and environmental factors. The change of plantar pressure in diabetic patients is one of the important reasons for the occurrence of diabetic foot. The abnormal increase of plantar pressure is a predictor of the common occurrence of foot ulcers. The feature extraction of plantar pressure distribution will be beneficial to the design and manufacture of diabetic shoes that will be beneficial for early protection of diabetes mellitus patients. In this research, texture-based features of the angular second moment (ASM), moment of inertia (MI), inverse difference monument (IDM), and entropy (E) have been selected and fused by using the up–down algorithm. The fused features are normalized to predict comfort plantar pressure imaging dataset using an improved fuzzy hidden Markov model (FHMM). In FHMM, type-I fuzzy set is proposed and fuzzy Baum–Welch algorithm is also applied to estimate the next features. The results are discussed, and by comparing with other back–forward algorithms and different fusion operations in FHMM. Improved HMMs with up–down fusion using type-I fuzzy definition performs high effectiveness in prediction comfort plantar pressure distribution in an image dataset with an accuracy of 82.2% and the research will be applied to the shoe-last personalized customization in the industry.

An efficient local binary pattern based plantar pressure optical sensor image classification using convolutional neural networks

The objective of this study was to design and produce highly comfortable shoe products guided by a plantar pressure imaging data-set. In this research a plantar pressure optical imaging data-set based classification technology has been developed. In this paper, an improved local binary pattern (LBP) algorithm is used to retrieve texture-based features and recognize patterns from the data-set. A calculation model of plantar pressure imaging feature area is established subsequently. The data-set is classified by a neural network to guide the generation of various shoe-last surfaces. Firstly, the local binary mode is improved to adapt to the pressure imaging data-set, and the texture-based feature calculation is fully used to accurately generate the feature point set; hereafter, the plantar pressure imaging feature point set is then used to guide the design of last free surface forming. In the presented experiments of plantar imaging, multi-dimensional texture-based features and improved LBP features have been found by a convolution neural network (CNN), and compared with a 21-input-3-output two-layer perceptual neural network. Three feet types are investigated in the experiment, being flatfoot (F) referring to the lack of a normal arch, or arch collapse, Talipes Equinovarus (TE), being the front part of the foot is adduction, calcaneus varus, plantar flexion, or Achilles tendon contracture and Normal (N). This research has achieved an 82% accuracy rate with 10 hidden-layers CNN of rotation invariance LBP (RI-LBP) algorithm using 21 texture-based features by comparing other deep learning methods presented in the literature.

A convolutional neural network Cascade for plantar pressure images registration

BackgroundPlantar pressure image (PPI) recorded in high spatial and temporal resolution is very useful in clinical gait analysis. For functional analysis of PPI, image registration is often performed to maximally correlate source image with a template image. Previous methods estimate the registration parameters by iteratively optimizing different objective functions. These methods are often computational expensive to achieve satisfactory registration accuracy. Research questionCan we develop a single PPI registration technique that performs more rapidly than previous methods, and that also maintains adequate PPI correspondence as defined by various (dis)similarity metrics? MethodsA cascaded convolutional neural network (CNN) was proposed for the registration of PPIs. Our model was trained to learn a regression from the difference between the template and misaligned images to the registration parameters. The registration performance was evaluated by three different metrics, i.e. the mean squared error (MSE), the exclusive or (XOR), and the mutual information (MI). For comparison, four previous methods were also implemented. These included the principal axes (PA) method, the center of pressure trajectory (COP) method, the MSE method, and the XOR method. ResultsExperimental results on a dataset with 71 PPI template-source pairs showed that the proposed CNN-based method could obtain comparable registration accuracy to the MSE and XOR method. With regards to the registration speed, registration durations (mean ± sd in seconds) per image pair were: MSE (30.584 ± 2.171), XOR (24.245 ± 1.596), PA (0.016 ± 0.003), COP (25.614 ± 0.341), and the proposed model (0.054 ± 0.007). SignificanceOur findings indicate that the proposed registration approach can achieve high accuracy but less computational time. Thus, it is more practical to utilize our pre-trained CNN-based model to develop near-real time applications for plantar pressure images registration.

Research on feature extraction algorithm for plantar pressure image and gait analysis in stroke patients

The plantar pressure image is an important tool for gait analysis. It has important applications in evaluating the recovery of stroke patients after operation and formulating the rehabilitation training program. It is one of the key technologies of gait analysis to extract foot feature parameters from static/dynamic plantar pressure images. This article deals with the noise in the original image through the piecewise linear grayscale transformation, the time domain mean filter and the maximum value filter, then determine the position of the feet in the image by the foot localization algorithm based on the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) and the K-means clustering method. Finally, the plantar pressure feature parameters were extracted according to the positioned images. Based on the above feature parameter extraction algorithm, the plantar pressure feature parameters of 20 healthy subjects and 20 S patients with relative recovery period (2–6 months after the onset) were compared, showing a statistically significant difference (P < 0.001). Based on the above data, gait characteristics of stroke patients were further analyzed.

RETRACTED ARTICLE: Evaluation on diabetic plantar pressure data-set employing auto-segmentation technologies

Plantar pressure imaging technologies play more important role in diabetics’ shoe-last design. The plantar pressure imaging data-set was acquired through pressure sensors system and was preprocessed using typical image process technologies including Gauss filtering, gamma correction, and wavelet transform enhancement in this work. To decrease the computational complexity, edge detection operator with Sobel, Roberts, Prewitt, Log and Canny were applied. Finally, threshold-based method, gray, watershed, feature clustering and fuzzy cluster, region growing-based plantar pressure image segmentation were employed, respectively. Results illustrated that threshold performs the better effectiveness by using analytic hierarchy process evaluation method through a special indices definition. The proposed methods in the research will be potential application and guidance for comfort shoe design for diabetics.

Optical pressure sensors based plantar image segmenting using an improved fully convolutional network

Optical pressure sensors in foot scanner system (FSS) based imaging systems can directly and accurately reflect the elastic changes through the plantar tissue to form different pressure values in different regions. In this work, the computational complexity of the sensor dataset from FSS was reduced using an improved full convolution network (FCN) through the AlexNet platform (FCN-AlexNet-8 s). Initially, plantar pressure imaging pre-process and segmentation techniques were developed to extract the region of interests (ROIs) in the captured images from the optical pressure sensors. The experiment established superior performance in the index evaluation system, including mean square error (MSE), and peak signal to noise ratio (PSNR) compared to the previous related studies. Furthermore, the proposed method was compared with region based convolutional neural network (R-CNN) and fast R-CNN separately in terms of the layers indices, maximum stride and time consuming. Accordingly, the proposed method is beneficial to decrease the computation complexity of plantar pressure sensor datasets and has potential application on shoe-last customization.

Diabetic plantar pressure analysis using image fusion

Plantar pressure images analysis is the key issue of designing comfortable shoe products through last customizing system, which has attracted the researchers’ curiosity toward image fusion as an application of medical and industrial imaging. In the current work, image fusion has been applied using wavelet transform and compared with Laplace Pyramid. Using image fusion rules of Mean-Max, we presented a plantar pressure image fusion method employing haar wavelet transform. It was compared in different composition layers with the Laplace pyramid transform. The experimental studies deployed the haar, db2, sym4, coif2, and bior5.5 wavelet basis functions for image fusion under decomposition layers of 3, 4, and 5. Evaluation metrics were measured in the case of the different layer number of wavelet decomposition to determine the best decomposition level and to evaluate the fused image quality using with different wavelet functions. The best wavelet basis function and decomposition layers were selected through the analysis and the evaluation measurements. This study established that haar wavelet transform with five decomposition levels on plantar pressure image achieved superior performance of 89.2817% mean, 89.4913% standard deviation, 5.4196 average gradient, 14.3364 spatial frequency, 5.9323 information entropy and 0.2206 cross entropy.

Histogram of Oriented Gradient Based Plantar Pressure Image Feature Extraction and Classification Employing Fuzzy Support Vector Machine

To identify and design comfortable footwear for patients with diabetic conditions (e.g. high blood pressure), an experimental design was proposed to collect plantar pressure data-set using a RSscan sensor system. The data-set acquired by the pressure sensor was reformed into images to allow for image analysis technologies to be applied. In this paper, image features were extracted including color of Hue-Saturation-Value (HSV), gray difference based features (mean, entropy and auto-correlation function), gray-level co-occurrence matrix based features (energy and correlation) and Histogram of Oriented Gradient (HOG). The features were normalized into a high dimensional vector applied to a Fuzzy Support Vector Machine (FSVM), and finally, the FSVM was trained and used for prediction of diabetic plantar pressure images. Normal features and HOG were compared in different classifiers including SVM, LSVM and FSVM. HOG with normal features of image for FSVM performed with a higher accuracy classification effectiveness (84.3%) than the current state of the art. The proposed methods have clear applications in revealing the key zone of foot plantar of diabetics and offer a new direction in producing comfortable diabetic footwear.

Convolutional Neural Network Based Clustering and Manifold Learning Method for Diabetic Plantar Pressure Imaging Dataset

Convolutional Neural Network Based Clustering and Manifold Learning Method for Diabetic Plantar Pressure Imaging Dataset

Eigenfoot decomposition of plantar pressure images and case study of feature prediction of two modalities.

The registration of plantar pressure images is a widely used technique to support human gait analysis. In plantar pressure images, most of the time conventionally derived features are used for further processing. Recently, automatic feature extraction based on PCA and kPCA is being used, to increase the information that can be extracted from this data. In this paper, we describe our work flow and a case study on the application of predicting two pressure features and a non-pressure feature out of the automatically derived PCA features. This includes the normalization of the pressure images, the PCA based feature extraction, and building and testing the regression model based on a linear and kernel SVM.

A knowledge-based modeling for plantar pressure image reconstruction.

It is known that prolonged pressure on the plantar area is one of the main factors in developing foot ulcers. With current technology, electronic pressure monitoring systems can be placed as an insole into regular shoes to continuously monitor the plantar area and provide evidence on ulcer formation process as well as insight for proper orthotic footwear design. The reliability of these systems heavily depends on the spatial resolution of their sensor platforms. However, due to the cost and energy constraints, practical wireless in-shoe pressure monitoring systems have a limited number of sensors, i.e., typically K < 10. In this paper, we present a knowledge-based regression model (SCPM) to reconstruct a spatially continuous plantar pressure image from a small number of pressure sensors. This model makes use of high-resolution pressure data collected clinically to train a per-subject regression function. SCPM is shown to outperform all other tested interpolation methods for K < 60 sensors, with less than one-third of the error for K = 10 sensors. SCPM bridges the gap between the technological capability and medical need and can play an important role in the adoption of sensing insole for a wide range of medical applications.

Enhanced spatio-temporal alignment of plantar pressure image sequences using B-splines

This article presents an enhanced methodology to align plantar pressure image sequences simultaneously in time and space. The temporal alignment of the sequences is accomplished using B-splines in the time modeling, and the spatial alignment can be attained using several geometric transformation models. The methodology was tested on a dataset of 156 real plantar pressure image sequences (3 sequences for each foot of the 26 subjects) that was acquired using a common commercial plate during barefoot walking. In the alignment of image sequences that were synthetically deformed both in time and space, an outstanding accuracy was achieved with the cubic B-splines. This accuracy was significantly better (p<0.001) than the one obtained using the best solution proposed in our previous work. When applied to align real image sequences with unknown transformation involved, the alignment based on cubic B-splines also achieved superior results than our previous methodology (p<0.001). The consequences of the temporal alignment on the dynamic center of pressure (COP) displacement was also assessed by computing the intraclass correlation coefficients (ICC) before and after the temporal alignment of the three image sequence trials of each foot of the associated subject at six time instants. The results showed that, generally, the ICCs related to the medio-lateral COP displacement were greater when the sequences were temporally aligned than the ICCs of the original sequences. Based on the experimental findings, one can conclude that the cubic B-splines are a remarkable solution for the temporal alignment of plantar pressure image sequences. These findings also show that the temporal alignment can increase the consistency of the COP displacement on related acquired plantar pressure image sequences.

RECENT METHODS FOR PLANTAR PRESSURE IMAGE REGISTRATION

Image registration is the process of aligned two or more images in a common coordinate system, in a way that similar structures become optimally overlapped. The registration of plantar pressure images has increased in the last years, as various time consuming tasks can be accurately and rapidly reached if the original images are efficiently registered. The aim of the present abstract is to introduce the image registration methods that we have been developed to register plantar pressure images.

Registration of plantar pressure images.

In this work, five computational methodologies to register plantar pressure images are compared: (1) the first methodology is based on matching the external contours of the feet; (2) the second uses the phase correlation technique; (3) the third addresses the direct maximization of cross-correlation using the Fourier transform; (4) the fourth minimizes the sum of squared differences using the Fourier transform; and (5) the fifth methodology iteratively optimizes an intensity (dis)similarity measure based on Powell's method. The accuracy and robustness of the five methodologies were assessed by using images from three common plantar pressure acquisition devices: a Footscan system, an EMED system, and a light reflection system. Using the residual error as a measure of accuracy, all methodologies revealed to be very accurate even in the presence of noise. The most accurate was the methodology based on the iterative optimization, when the mean squared error was minimized. It achieved a residual error inferior to 0.01 mm and 0.6 mm for non-noisy and noisy images, respectively. On the other hand, the methodology based on image contour matching was the fastest, but its accuracy was the lowest.

Towards an efficient and robust foot classification from pedobarographic images

This paper presents a new computational framework for automatic foot classification from digital plantar pressure images. It classifies the foot as left or right and simultaneously calculates two well-known footprint indices: the Cavanagh's arch index (AI) and the modified AI. The accuracy of the framework was evaluated using a set of plantar pressure images from two common pedobarographic devices. The results were outstanding, as all feet under analysis were correctly classified as left or right and no significant differences were observed between the footprint indices calculated using the computational solution and the traditional manual method. The robustness of the proposed framework to arbitrary foot orientations and to the acquisition device was also tested and confirmed.

Computer-Based Identification of Type 2 Diabetic Subjects with and Without Neuropathy Using Dynamic Planter Pressure and Principal Component Analysis

Diabetes is a chronic disease that is characterized by an increased blood glucose level due to insulin resistance. Type 2 diabetes is common in middle aged and old people. In this work, we present a technique to analyze dynamic foot pressures images and classify them into normal, diabetes type 2 with and without neuropathy classes. Plantar pressure images were obtained using the F-Scan (Tekscan, USA) in-shoe measurement system. We used Principal Component Analysis (PCA) and extracted the eigenvalues from different regions of the foot image. The features extracted from region 1 of the foot pressure image, which were found to be clinically significant, were fed into the Fuzzy classifier (Sugeno model) for automatic classification. Our results show that the proposed method is able to identify the unknown class with an accuracy of 93.7%, sensitivity of 100%, and specificity of 83.3%. Moreover, in this work, we have proposed an integrated index using the eigenvalues to differentiate the normal subjects from diabetes with and without neuropathy subjects using just one number. This index will help the clinicians in easy and objective daily screening, and it can also be used as an adjunct tool to cross check their diagnosis.

Spatio-temporal alignment of pedobarographic image sequences

This article presents a methodology to align plantar pressure image sequences simultaneously in time and space. The spatial position and orientation of a foot in a sequence are changed to match the foot represented in a second sequence. Simultaneously with the spatial alignment, the temporal scale of the first sequence is transformed with the aim of synchronizing the two input footsteps. Consequently, the spatial correspondence of the foot regions along the sequences as well as the temporal synchronizing is automatically attained, making the study easier and more straightforward. In terms of spatial alignment, the methodology can use one of four possible geometric transformation models: rigid, similarity, affine, or projective. In the temporal alignment, a polynomial transformation up to the 4th degree can be adopted in order to model linear and curved time behaviors. Suitable geometric and temporal transformations are found by minimizing the mean squared error (MSE) between the input sequences. The methodology was tested on a set of real image sequences acquired from a common pedobarographic device. When used in experimental cases generated by applying geometric and temporal control transformations, the methodology revealed high accuracy. In addition, the intra-subject alignment tests from real plantar pressure image sequences showed that the curved temporal models produced better MSE results (P<0.001) than the linear temporal model. This article represents an important step forward in the alignment of pedobarographic image data, since previous methods can only be applied on static images.

Novel framework for registration of pedobarographic image data

This article presents a framework to register (or align) plantar pressure images based on a hybrid registration approach, which first establishes an initial registration that is subsequently improved by the optimization of a selected image (dis)similarity measure. The initial registration has two different solutions: one based on image contour matching and the other on image cross-correlation. In the final registration, a multidimensional optimization algorithm is applied to one of the following (dis)similarity measures: the mean squared error (MSE), the mutual information, and the exclusive or (XOR). The framework has been applied to intra- and inter-subject registration. In the former, the framework has proven to be extremely accurate and fast (<70ms on a normal PC notebook), and obtained superior XOR and identical MSE values compared to the best values reported in previous studies. Regarding the inter-subject registration, by using rigid, similarity, affine, projective, and polynomial (up to the fourth degree) transformations, the framework significantly optimized the image (dis)similarity measures. Thus, it is considered to be very accurate, fast, and robust in terms of noise, as well as being extremely versatile, all of which are regarded as essential features for near-real-time applications.