Two-level Decision Tree Research Articles

Uncertainty-driven ensembles of multi-scale deep architectures for image classification

The use of automatic systems for medical image classification has revolutionized the diagnosis of a high number of diseases. These alternatives, which are usually based on artificial intelligence (AI), provide a helpful tool for clinicians, eliminating the inter and intra-observer variability that the diagnostic process entails. Convolutional Neural Network (CNNs) have proved to be an excellent option for this purpose, demonstrating a large performance in a wide range of contexts. However, it is also extremely important to quantify the reliability of the model’s predictions in order to guarantee the confidence in the classification. In this work, we propose a multi-level ensemble classification system based on a Bayesian Deep Learning approach in order to maximize performance while providing the uncertainty of each classification decision. This tool combines the information extracted from different architectures by weighting their results according to the uncertainty of their predictions. Performance is evaluated in a wide range of real scenarios: in the first one, the aim is to differentiate between different pulmonary pathologies: controls vs bacterial pneumonia vs viral pneumonia. A two-level decision tree is employed to divide the 3-class classification into two binary classifications, yielding an accuracy of 98.19%. In the second context, performance is assessed for the diagnosis of Parkinson’s disease, leading to an accuracy of 95.31%. The reduced preprocessing needed for obtaining this high performance, in addition to the information provided about the reliability of the predictions evidence the applicability of the system to be used as an aid for clinicians.

A radiomics model can distinguish solitary pulmonary capillary haemangioma from lung adenocarcinoma.

OBJECTIVESSolitary pulmonary capillary haemangioma (SPCH) is a benign lung tumour that presents as ground-glass nodules on computed tomography (CT) images and mimics lepidic-predominant adenocarcinoma. This study aimed to establish a discriminant model using a radiomic feature analysis to distinguish SPCH from lepidic-predominant adenocarcinoma.METHODSIn the adenocarcinoma group, all tumours were of the lepidic-predominant subtype with high purity (>70%). A classification model was proposed based on a two-level decision tree and 26 radiomic features extracted from each segmented lesion. For comparison, a baseline model was built with the same 26 features using a support vector machine as the classifier. Both models were assessed by the leave-one-out cross-validation method.RESULTSThis study included 13 and 49 patients who underwent complete resection for SPCH and adenocarcinoma, respectively. Two sets of features were identified for discrimination between the 2 different histology types. The first set included 2 principal components corresponding to the 2 largest eigenvalues for the root node of the two-level decision tree. The second set comprised 4 selected radiomic features. The area under the receiver operating characteristic curve, accuracy, sensitivity, specificity were 0.954, 91.9%, 92.3% and 91.8% in the proposed classification model, and were 0.805, 85.5%, 61.5% and 91.8% in the baseline model, respectively. The proposed classification model significantly outperformed the baseline model (P < 0.05).CONCLUSIONSThe proposed model could differentiate the 2 different histology types on CT images, and this may help surgeons to preoperatively discriminate SPCH from adenocarcinoma.

WiFi CSI-Based Human Activity Recognition Using Deep Recurrent Neural Network

Human activity recognition based on channel state information (CSI) using commercial WiFi devices plays an increasingly important role in many applications, such as smart home and interactive games. In this paper, we propose a WiFi CSI based human activity recognition approach using deep recurrent neural network (HARNN). HARNN mainly exploits four key techniques to recognize different human activities. HARNN firstly constructs a novel two-level decision tree for using two environment variation statistics efficiently. Meanwhile, a linear regression method is also introduced to seek for the optimal parameter for the designed decision tree. Depending on this, the decision tree is used to sense indoor environment variation, and then detect whether there is any human activity occurring in a target area. In addition, a noise removal mechanism is devised to eliminate the influence of random noise derived from indoor environments. Then, to characterize various human activities, two representative features are extracted from different statistical profiles, including channel power variation (CPV) and time-frequency analysis (TFA). Finally, a recurrent neural network (RNN) model is utilized to recognize different human activities by leveraging the extracted representative features above. According to the above steps, the proposed HARNN could establish a robust relationship between human activities and WiFi CSI compared with most of the existing WiFi CSI based approaches. The proof-of-concept prototype of HARNN is implemented on a set of commercial WiFi devices, and its overall performance is evaluated in several typical indoor environments. The experimental results demonstrate that HARNN can achieve better recognition performance compared with some benchmark approaches.

Optimized pre-processing input plane GPU implementation of an optical face recognition technique using a segmented phase only composite filter

The key outcome of this work is to propose and validate a fast and robust correlation scheme for face recognition applications. The robustness of this fast correlator is ensured by an adapted pre-processing step for the target image allowing us to minimize the impact of its (possibly noisy and varying) amplitude spectrum information. A segmented composite filter is optimized, at the very outset of its fabrication, by weighting each reference with a specific coefficient which is proportional to the occurrence probability. A hierarchical classification procedure (called two-level decision tree learning approach) is also used in order to speed up the recognition procedure. Experimental results validating our approach are obtained with a prototype based on GPU implementation of the all-numerical correlator using the NVIDIA GPU GeForce 8400GS processor and test samples from the Pointing Head Pose Image Database (PHPID), e.g. true recognition rates larger than 85% with a run time lower than 120ms have been obtained using fixed images from the PHPID, true recognition rates larger than 77% using a real video sequence with 2 frame per second when the database contains 100 persons. Besides, it has been shown experimentally that the use of more recent GPU processor like NVIDIA-GPU Quadro FX 770M can perform the recognition of 4 frame per second with the same length of database.

Determination of Grade and Subtype of Meningiomas by Using Histogram Analysis of Diffusion-Tensor Imaging Metrics

To determine whether histogram analysis of diffusion-tensor (DT) magnetic resonance (MR) imaging metrics, including tensor shape measurements, can help determine the grades and subtypes of meningiomas. The institutional review board approved this HIPAA-compliant study. Nine atypical, three anaplastic, and 39 typical meningiomas were retrospectively studied. The 39 typical meningiomas included one secretory meningioma and 11 fibroblastic, 11 transitional, 14 meningothelial, and two angiomatous meningiomas. DT imaging metrics, including fractional anisotropy, mean diffusivity, linear anisotropy coefficient, planar anisotropy coefficient (CP), spherical anisotropy coefficient (CS), and eigenvalue skewness (SK), as well as normalized signal intensity from contrast-enhanced T1- and T2-weighted images, were measured from the enhancing region of the tumor. Mean, variance, skewness, and kurtosis were extracted from the histograms. A two-level decision tree was designed, and a multivariate logistic regression analysis was used at each level to determine the best model for classification. Histogram skewness of SK and kurtosis of SK were significantly higher in atypical and anaplastic meningiomas than in typical meningiomas (P<.01). Among typical meningiomas, significant differences in histogram measures of CP and CS between fibroblastic meningiomas and other subtypes were observed (P<.01). The best model for differentiating atypical and anaplastic meningiomas from typical meningiomas consisted of mean and skewness of SK and kurtosis of T1 signal intensity, with an area under the receiver operating characteristic curve (AUC) of 0.946. The best model for differentiating fibroblastic meningiomas from other subtypes consisted of skewness of T2 signal intensity and kurtosis of CP (AUC, 0.970). Histogram analysis of DT imaging metrics can help determine the grades and subtypes of meningiomas, which can better assist in surgical planning.

Development of two-level decision tree-based real-time scheduling system under product mix variety environment

Most of the research on machine learning-based real-time scheduling (RTS) systems has been aimed toward product constant mix environments. However, in a product mix variety manufacturing environment, the scheduling knowledge base (KB) is dynamic; therefore, it would be interesting to develop a procedure that would automatically modify the scheduling knowledge when important changes occur in the manufacturing system. All of the machine learning-based RTS systems (including a KB refinement mechanism) proposed in earlier studies periodically require the addition of new training samples and regeneration of new KBs. Hence, previous approaches investigating machine learning-based RTS systems have been confronted with the training data overflow problem and an increase in the scheduling KB building time, which are unsuitable for RTS control. The objective of this paper is to develop a KB class selection mechanism that can be supported in various product mix ratio environments. Hence, the RTS KB is developed by a two-level decision tree (DT) learning approach. First, a suitable scheduling KB class is selected. Then, for each KB class, the best (proper) dispatching rule is selected for the next scheduling period. Here, the proposed two-level DT RTS system comprises five key components: (1) training samples generation mechanism, (2) GA/DT-based feature selection mechanism, (3) building a KB class label by a two-level self-organizing map, (4) DT-based KB class selection module, and (5) DT-based dynamic dispatching rule selection module. The proposed two-level DT-based KB RTS system yields better system performance than that by a one-level DT-based RTS system and heuristic individual dispatching rules in a flexible manufacturing system under various performance criteria over a long period.

Guideline-based modeling of therapeutic strategies in the special case of chronic diseases

Despite the availability of evidence-based clinical practice guidelines in most countries, patients with chronic diseases are still generally inadequately managed. One difficulty lies in the optimal synchronization of a patient with the guideline therapeutic strategy, especially when the history of her past treatments does not follow the recommended sequence of therapies. We propose a formal model to represent guideline-based therapeutic strategies as a two-level decision tree. The clinical level is used to identify a patient-specific clinical situation, on the basis of key elements of clinical examination (complication of hypertension, associated diseases). The therapeutic level is derived from the formalization of guideline-based strategies first represented as bidimensional matrices structured in lines of therapy and levels of therapeutic intention. A revised version based on the ordering of levels of therapeutic combination is then developed. The aim is to dynamically provide the best next step of treatment from the patient's therapeutic-history-based customization of the general therapeutic sequence established in the guideline for the corresponding clinical situation. A preliminary in vitro evaluation of the system on actual clinical cases showed a positive impact on physician compliance with a significant increase from 16 to 57%.

Disaggregation of polygons of surficial geology and soil maps using spatial modelling and legacy data

Examples from the Murray–Darling basin in Australia are used to illustrate different methods of disaggregation of reconnaissance-scale maps. One approach for disaggregation revolves around the de-convolution of the soil–landscape paradigm elaborated during a soil survey. The descriptions of soil map units and block diagrams in a soil survey report detail soil–landscape relationships or soil toposequences that can be used to disaggregate map units into component landscape elements. Toposequences can be visualised on a computer by combining soil maps with digital elevation data. Expert knowledge or statistics can be used to implement the disaggregation. Use of a restructuring element and k-means clustering are illustrated. Another approach to disaggregation uses training areas to develop rules to extrapolate detailed mapping into other, larger areas where detailed mapping is unavailable. A two-level decision tree example is presented. At one level, the decision tree method is used to capture mapping rules from the training area; at another level, it is used to define the domain over which those rules can be extrapolated.

Typeface Identification for Printed Chinese Characters

In this paper, we propose a methodology for identifying typefaces of printed Chinese characters in documents. Three kinds of features, stroke width means, stroke width variations, and aspect ratio, are first used to classify character typefaces as: Black, Li, Kai-Round, or Ming-Song. Each of the last two groups contains two typefaces. Vertical/horizontal stroke width ratios are used to distinguish between the Ming and Song typefaces and accumulative pixel ratio to distinguish between the Kai and Round typefaces. Six different typeface feature distributions measured from 5401 printed Chinese characters are considered, and a trapezoid-shaped membership function is constructed for each distribution. Based on these membership functions, we determine what typeface each input character belongs to using a two-level decision tree. To increase the identification rate, the typeface of a certain character is adjusted according to the typeface identification results of the front and the next characters. In the character recognition system, we use two statistical features: crossing counts and contour directional counts. We achieved an 89.87% typeface identification rate in our experiments, and a 95.60% character recognition rate.