Alternating Decision Tree Research Articles

GIS-based landslide susceptibility modelling: a comparative assessment of kernel logistic regression, Naïve-Bayes tree, and alternating decision tree models

ABSTRACTThe main purpose of this paper is to explore some potential applications of sophisticated machine learning techniques such as the kernel logistic regression, Naïve-Bayes tree and alternating decision tree models for landslide susceptibility analysis at Taibai county (China). Initially, a landslide inventory map containing the information of 212 historical landslide locations was prepared. Seventy percentage (148) of landslides were randomly selected for training models and the remaining were used for validation. Additionally, 12 landslide conditioning factors were considered and the thematic layers were prepared in GIS. Subsequently, these three models were applied to build landslide susceptibility maps. The performances of the models were compared using the receive operating characteristic curves, kappa index, and statistical evaluation measures. The results show that the KLR model has the highest AUC values of 0.910 and 0.936 for training and validation datasets, respectively. The KLR model also has the highest degree of goodness-of-fits (84.5%) for the training dataset. The NBTree model has the highest goodness-of-fits (91.4%) for the validation dataset. However, the KLR model has the preferable balance performance for both the training and validation process. The results of this study demonstrate the benefit of selecting the optimal machine learning techniques in landslide susceptibility mapping.

A practical decision-tree model to predict complexity of reconstructive surgery after periocular basal cell carcinoma excision.

Periocular basal cell carcinomas (pBCC) have unpredictable growth. The authors seek to derive a decision rule for predicting surgical complexity in pBCC. This study was conducted at two centres in New Zealand from September 2010 to November 2015. Baseline demographic information and an initial assessment of operative complexity (a four-point grading scale) were collected. Assessment of operative complexity was repeated at the time of reconstruction. Univariate analysis was applied to identify the associative factors and supervised machine learning was used to determine the best predictive models to construct a clinical decision rule. A total of 156 patients and 156 periocular BCC were analysed. Univariate analysis revealed that older age, recurrent skin cancer, large tumour size, being a public patient and high complexity at pre-operative assessment were associated with high actual operative complexity. Tumour histology was not associated with more complex surgery. Machine learning analyses revealed that Naive Bayesian classifier was able to distinguish surgical complexity with an average area under the receiver operating characteristic curve (AUC) of 0.854 (95% CI: 0.762-0.946) whereas a simpler, alternating decision tree (ADT) that used only three clinical variables achieved an AUC of 0.853 (95% CI: 0.739-0.931). The ADT model was 10.1 times more likely to correctly identify a high complexity case. The three predictive variables were pre-operative assessment of complexity (high vs. low), surgical delays [early (<75 days) or delayed (≥75 days)], and tumour size [small (<14 mm), or large (≥14 mm)]. For the subgroup with large tumours but low initial assessed complexity, late surgery was associated with a 6.7-fold increase in risk of high-risk surgery. A simple, three-variable risk stratification system was able to predict the operative complexity of pBCC.

Predictive value of CD24 and CD44 for neoadjuvant chemotherapy response and prognosis in primary breast cancer patients.

We investigated the significance of CD24 and CD44 expression for predicting responses to chemotherapy and prognosis in primary breast cancer patients. Diagnosis of breast cancer was confirmed by core needle biopsy, and immunohistochemical studies were performed. Preoperatively, patients received anthracycline-containing chemotherapy. Expression of CD44 and CD24 was assessed immunohistochemically and the relationship with chemotherapy response and with prognosis was analyzed. Between 2001 and 2004, 139 women were enrolled in this study. In the correlation analysis, CD24 expression was negatively associated with pathological response to chemotherapy (p = 0.0003). A machine learning technique with an alternating decision tree (ADTree) showed that four logical rules are involved in predicting the response depending on the combination of CD24, HER2, tumor stage, CD44, progesterone receptor, and patient age. In the survival analysis, patients having CD44 (++) showed a significantly favorable prognosis as compared with others (p = 0.0002). A multivariate analysis showed that CD44 expression had an independent prognostic value (p < .0001). We found a significant correlation between CD44 expression and prognosis and between CD24 expression and response to chemotherapy. CD24 and CD44 expressions would be useful predictive markers, although further studies are needed.

A Comparative Study of Least Square Support Vector Machines and Multiclass Alternating Decision Trees for Spatial Prediction of Rainfall-Induced Landslides in a Tropical Cyclones Area

The objective of this study is to explore and compare the least square support vector machine (LSSVM) and multiclass alternating decision tree (MADT) techniques for the spatial prediction of landslides. The Luc Yen district in Yen Bai province (Vietnam) has been selected as a case study. LSSVM and MADT are effective machine learning techniques of classification applied in other fields but not in the field of landslide hazard assessment. For this, Landslide inventory map was first constructed with 95 landslide locations identified from aerial photos and verified from field investigations. These landslide locations were then divided randomly into two parts for training (70 % locations) and validation (30 % locations) processes. Secondly, landslide affecting factors such as slope, aspect, elevation, curvature, lithology, land use, distance to roads, distance to faults, distance to rivers, and rainfall were selected and applied for landslide susceptibility assessment. Subsequently, the LSSVM and MADT models were built to assess the landslide susceptibility in the study area using training dataset. Finally, receiver operating characteristic curve and statistical index-based evaluations techniques were employed to validate the predictive capability of these models. As a result, both the LSSVM and MADT models have high performance for spatial prediction of landslides in the study area. Out of these, the MADT model (AUC = 0.853) outperforms the LSSVM model (AUC = 0.803). From the landslide study of Luc Yen district in Yen Bai province (Vietnam), it can be conclude that the LSSVM and MADT models can be applied in other areas of world also for and spatial prediction. Landslide susceptibility maps obtained from this study may be helpful in planning, decision making for natural hazard management of the areas susceptible to landslide hazards.

Predictability of Pressure Ulcers Based on Operation Duration, Transfer Activity, and Body Mass Index Through the Use of an Alternating Decision Tree.

To develop a prediction model for pressure ulcer cases that continue to occur at an acute care hospital with a low occurrence rate of pressure ulcers. Analyzing data were collected from patients hospitalized at Tokushima University Hospital during 2012 using an alternating decision tree (ADT) data mining method. The ADT-based analysis revealed transfer activity, operation time, and low body mass index (BMI) as important factors for predicting pressure ulcer development. Among the factors identified, only "transfer activity" can be modified by nursing intervention. While shear force and friction are known to lead to pressure ulcers, transfer activity has not been identified as such. Our results suggest that transfer activities creating shear force and friction correlate with pressure ulcer development. The ADT algorithm was effective in determining prediction factors, especially for highly imbalanced data. Our three stumps ADT yielded accuracy, sensitivity, and specificity values of 72.1%±3.7%, 79.3%±18.1%, and 72.1%±3.8%, respectively. Transfer activity, identified as an interventional factor, can be modified through nursing interventions to prevent pressure ulcer formation. The ADT method was effective in identifying factors within largely imbalanced data. J. Med. Invest. 63: 248-255, August, 2016.

Accurate Prediction of Advanced Liver Fibrosis Using the Decision Tree Learning Algorithm in Chronic Hepatitis C Egyptian Patients.

Background/Aim. Respectively with the prevalence of chronic hepatitis C in the world, using noninvasive methods as an alternative method in staging chronic liver diseases for avoiding the drawbacks of biopsy is significantly increasing. The aim of this study is to combine the serum biomarkers and clinical information to develop a classification model that can predict advanced liver fibrosis. Methods. 39,567 patients with chronic hepatitis C were included and randomly divided into two separate sets. Liver fibrosis was assessed via METAVIR score; patients were categorized as mild to moderate (F0–F2) or advanced (F3-F4) fibrosis stages. Two models were developed using alternating decision tree algorithm. Model 1 uses six parameters, while model 2 uses four, which are similar to FIB-4 features except alpha-fetoprotein instead of alanine aminotransferase. Sensitivity and receiver operating characteristic curve were performed to evaluate the performance of the proposed models. Results. The best model achieved 86.2% negative predictive value and 0.78 ROC with 84.8% accuracy which is better than FIB-4. Conclusions. The risk of advanced liver fibrosis, due to chronic hepatitis C, could be predicted with high accuracy using decision tree learning algorithm that could be used to reduce the need to assess the liver biopsy.

Multivariate alternating decision trees

Decision trees are comprehensible, but at the cost of a relatively lower prediction accuracy compared to other powerful black-box classifiers such as SVMs. Boosting has been a popular strategy to create an ensemble of decision trees to improve their classification performance, but at the expense of comprehensibility advantage. To this end, alternating decision tree (ADTree) has been proposed to allow boosting within a single decision tree to retain comprehension. However, existing ADTrees are univariate, which limits their applicability. This research proposes a novel algorithm – multivariate ADTree. It presents and discusses its different variations (Fisher׳s ADTree, Sparse ADTree, and Regularized Logistic ADTree) along with their empirical validation on a set of publicly available datasets. It is shown that multivariate ADTree has high prediction accuracy comparable to that of decision tree ensembles, while retaining good comprehension which is close to comprehension of individual univariate decision trees.

Identifying binge drinkers based on parenting dimensions and alcohol-specific parenting practices: building classifiers on adolescent-parent paired data

BackgroundMost Dutch adolescents aged 16 to 18 engage in binge drinking. Previous studies have investigated how parenting dimensions and alcohol-specific parenting practices are related to adolescent alcohol consumption. Mixed results have been obtained on both dimensions and practices, highlighting the complexity of untangling alcohol-related factors. The aim of this study was to investigate (1) whether parents’ reports of parenting dimensions and alcohol-specific parenting practices, adolescents’ perceptions of these dimensions and practices, or a combination are most informative to identify binge drinkers, and (2) which of these parenting dimensions and alcohol-specific parenting practices are most informative to identify binge drinkers.MethodsSurvey data of 499 adolescent-parent dyads were collected. The computational technique of data mining was used to allow for a data driven exploration of nonlinear relationships. Specifically, a binary classification task, using an alternating decision tree, was conducted and measures regarding the performance of the classifiers are reported after a 10-fold cross-validation.ResultsDepending on the parenting dimension or practice, parents’ reports correctly identified the drinking behaviour of 55.8 % (using psychological control) up to 70.2 % (using rules) of adolescents. Adolescents’ perceptions were best at identifying binge drinkers whereas parents’ perceptions were best at identifying non-binge drinkers.ConclusionsOf the parenting dimensions and practices, rules are particularly informative in understanding drinking behaviour. Adolescents’ perceptions and parents’ reports are complementary as they can help identifying binge drinkers and non-binge drinkers respectively, indicating that surveying specific aspects of adolescent-parent dynamics can improve our understanding of complex addictive behaviours.

Prediction of Allogeneic Hematopoietic Stem-Cell Transplantation Mortality 100 Days After Transplantation Using a Machine Learning Algorithm: A European Group for Blood and Marrow Transplantation Acute Leukemia Working Party Retrospective Data Mining Study.

Allogeneic hematopoietic stem-cell transplantation (HSCT) is potentially curative for acute leukemia (AL), but carries considerable risk. Machine learning algorithms, which are part of the data mining (DM) approach, may serve for transplantation-related mortality risk prediction. This work is a retrospective DM study on a cohort of 28,236 adult HSCT recipients from the AL registry of the European Group for Blood and Marrow Transplantation. The primary objective was prediction of overall mortality (OM) at 100 days after HSCT. Secondary objectives were estimation of nonrelapse mortality, leukemia-free survival, and overall survival at 2 years. Donor, recipient, and procedural characteristics were analyzed. The alternating decision tree machine learning algorithm was applied for model development on 70% of the data set and validated on the remaining data. OM prevalence at day 100 was 13.9% (n=3,936). Of the 20 variables considered, 10 were selected by the model for OM prediction, and several interactions were discovered. By using a logistic transformation function, the crude score was transformed into individual probabilities for 100-day OM (range, 3% to 68%). The model's discrimination for the primary objective performed better than the European Group for Blood and Marrow Transplantation score (area under the receiver operating characteristics curve, 0.701 v 0.646; P<.001). Calibration was excellent. Scores assigned were also predictive of secondary objectives. The alternating decision tree model provides a robust tool for risk evaluation of patients with AL before HSCT, and is available online (http://bioinfo.lnx.biu.ac.il/∼bondi/web1.html). It is presented as a continuous probabilistic score for the prediction of day 100 OM, extending prediction to 2 years. The DM method has proved useful for clinical prediction in HSCT.

Spatial prediction of landslide hazard at the Yihuang area (China) using two-class kernel logistic regression, alternating decision tree and support vector machines

Preparation of landslide susceptibility map is the first step for landslide hazard mitigation and risk assessment. The main aim of this study is to explore potential applications of two new models such as two-class Kernel Logistic Regression (KLR) and Alternating Decision Tree (ADT) for landslide susceptibility mapping at the Yihuang area (China). The ADT has not been used in landslide susceptibility modeling and this paper attempts a novel application of this technique. For the purpose of comparison, a conventional method of Support Vector Machines (SVM) which has been widely used in the literature was included and their results were assessed. At first, a landslide inventory map with 187 landslide locations for the study area was constructed from various sources. Landslide locations were then spatially randomly split in a ratio of 70/30 for building landslide models and for the model validation. Then a spatial database with a total of fourteen landslide conditioning factors was prepared, including slope, aspect, altitude, topographic wetness index (TWI), stream power index (SPI), sediment transport index (STI), plan curvature, landuse, normalized difference vegetation index (NDVI), lithology, distance to faults, distance to rivers, distance to roads, and rainfall. Using the KLR, the SVM, and the ADT, three landslide susceptibility models were constructed using the training dataset. The three resulting models were validated and compared using the receive operating characteristic (ROC), Kappa index, and five statistical evaluation measures. In addition, pairwise comparisons of the area under the ROC curve were carried out to assess if there are significant differences on the overall performance of the three models. The goodness-of-fits are 92.5% (the KLR model), 88.8% (the SVM model), and 95.7% (the ADT model). The prediction capabilities are 81.1%, 84.2%, and 93.3% for the KLR, the SVM, and the ADT models, respectively. The result shows that the ADT model yielded better overall performance and accurate results than the KLR and SVM models. The KLR model considered slightly better than SVM model in terms of the positive prediction values. The ADT and KLR are the two promising data mining techniques which might be considered to use in landslide susceptibility mapping. The results from this study may be useful for landuse planning and decision making in landslide prone areas.

Network-based prediction and knowledge mining of disease genes.

BackgroundIn recent years, high-throughput protein interaction identification methods have generated a large amount of data. When combined with the results from other in vivo and in vitro experiments, a complex set of relationships between biological molecules emerges. The growing popularity of network analysis and data mining has allowed researchers to recognize indirect connections between these molecules. Due to the interdependent nature of network entities, evaluating proteins in this context can reveal relationships that may not otherwise be evident.MethodsWe examined the human protein interaction network as it relates to human illness using the Disease Ontology. After calculating several topological metrics, we trained an alternating decision tree (ADTree) classifier to identify disease-associated proteins. Using a bootstrapping method, we created a tree to highlight conserved characteristics shared by many of these proteins. Subsequently, we reviewed a set of non-disease-associated proteins that were misclassified by the algorithm with high confidence and searched for evidence of a disease relationship.ResultsOur classifier was able to predict disease-related genes with 79% area under the receiver operating characteristic (ROC) curve (AUC), which indicates the tradeoff between sensitivity and specificity and is a good predictor of how a classifier will perform on future data sets. We found that a combination of several network characteristics including degree centrality, disease neighbor ratio, eccentricity, and neighborhood connectivity help to distinguish between disease- and non-disease-related proteins. Furthermore, the ADTree allowed us to understand which combinations of strongly predictive attributes contributed most to protein-disease classification. In our post-processing evaluation, we found several examples of potential novel disease-related proteins and corresponding literature evidence. In addition, we showed that first- and second-order neighbors in the PPI network could be used to identify likely disease associations.ConclusionsWe analyzed the human protein interaction network and its relationship to disease and found that both the number of interactions with other proteins and the disease relationship of neighboring proteins helped to determine whether a protein had a relationship to disease. Our classifier predicted many proteins with no annotated disease association to be disease-related, which indicated that these proteins have network characteristics that are similar to disease-related proteins and may therefore have disease associations not previously identified. By performing a post-processing step after the prediction, we were able to identify evidence in literature supporting this possibility. This method could provide a useful filter for experimentalists searching for new candidate protein targets for drug repositioning and could also be extended to include other network and data types in order to refine these predictions.

Sparse alternating decision tree

Alternating decision tree (ADTree) is a special decision tree representation that brings interpretability to boosting, a well-established ensemble algorithm. This has found success in wide applications. However, existing variants of ADTree are implementing univariate decision nodes where potential interactions between features are ignored. To date, there has been no multivariate ADTree. We propose a sparse version of multivariate ADTree such that it remains comprehensible. The proposed sparse ADTree is empirically tested on UCI datasets as well as spectral datasets from the University of Eastern Finland (UEF). We show that sparse ADTree is competitive against both univariate decision trees (original ADTree, C4.5, and CART) and multivariate decision trees (Fisher's decision tree and a single multivariate decision tree from oblique Random Forest). It achieves the best average rank in terms of prediction accuracy, second in terms of decision tree size and faster induction time than existing ADTree. In addition, it performs especially well on datasets with correlated features such as UEF spectral datasets. Thus, the proposed sparse ADTree extends the applicability of ADTree to a wider variety of applications.

Machine learning approaches for integrating clinical and imaging features in late-life depression classification and response prediction.

Currently, depression diagnosis relies primarily on behavioral symptoms and signs, and treatment is guided by trial and error instead of evaluating associated underlying brain characteristics. Unlike past studies, we attempted to estimate accurate prediction models for late-life depression diagnosis and treatment response using multiple machine learning methods with inputs of multi-modal imaging and non-imaging whole brain and network-based features. Late-life depression patients (medicated post-recruitment) (n = 33) and older non-depressed individuals (n = 35) were recruited. Their demographics and cognitive ability scores were recorded, and brain characteristics were acquired using multi-modal magnetic resonance imaging pretreatment. Linear and nonlinear learning methods were tested for estimating accurate prediction models. A learning method called alternating decision trees estimated the most accurate prediction models for late-life depression diagnosis (87.27% accuracy) and treatment response (89.47% accuracy). The diagnosis model included measures of age, Mini-mental state examination score, and structural imaging (e.g. whole brain atrophy and global white mater hyperintensity burden). The treatment response model included measures of structural and functional connectivity. Combinations of multi-modal imaging and/or non-imaging measures may help better predict late-life depression diagnosis and treatment response. As a preliminary observation, we speculate that the results may also suggest that different underlying brain characteristics defined by multi-modal imaging measures-rather than region-based differences-are associated with depression versus depression recovery because to our knowledge this is the first depression study to accurately predict both using the same approach. These findings may help better understand late-life depression and identify preliminary steps toward personalized late-life depression treatment.

Performance Analysis of Decision Tree Algorithms for Breast Cancer Classification

Backgrounds/Objectives: Data Mining (DM) techniques are extremely utilized for the extraction of useful information which is available in data warehouses and other database repositories. In medical diagnose, the role of DM approach rises quick recognition of disease over symptoms. To classify the medical data, a number of DM techniques are used by researchers. One of such techniques is classification. The classification algorithms predict the hidden information in the medical domain. The breast cancer is the very dangerous disease for women in developed countries like India. Most of the women death happens in the world, they are affected by the breast cancer. Methods/Statistical Analysis: The role of classification is importantin the real world applications in every field. Classification is used to classify the elements permitting to the features of the elements through the predefined set of classes. This research work analyses the breast cancer data using classification algorithms namely j48, Classification and Regression Trees (CART), Alternating Decision Tree (AD Tree) and Best First Tree (BF Tree). Findings: To find the performance of classification algorithms, this work uses cancer data as input. Particularly, this work is carried out to compare the four decision tree algorithms in the prediction of the performance accuracy in breast cancer data. All the algorithms are applied for breast cancer data to classify the data set for classification and prediction. Among these four methods, this work concludes the best algorithm for the chosen input data on decision tree supervised learning algorithms to predict the best classifier. Applications/Improvements: The breast cancer data is analyzed by taking the images using the same algorithms in future. Also, the microcalcifications of the breast cancer imagery are to be investigated in the same work. Keywords: CART Algorithm, Classification Algorithms, Decision Trees, J48 Algorithm

DATA MINING APPROACH FOR IN-HOSPITAL TREATMENT OUTCOME IN PATIENTS WITH ACUTE CORONARY SYNDROME.

Risk stratification is nowadays crucial when estimating the patient's prognosis in terms of treatment outcome and it also helps in clinical decision making. Several risk assessment models have been developed to predict short-term outcomes in patients with acute coronary syndrome. This study was aimed at developing an outcome prediction model for patients with acute coronary syndrome submitted to percutaneus coronary intervention using data mining approach. A total of 2030 patients hospitalized for acute coronary syndrome and treated with percutaneous coronary intervention from December 2008 to December 2011 were assigned to a derivation cohort. Demographic and anamnestic data, clinical characteristics on admission, biochemical analysis of blood parameters on admission, and left ventricular ejection fraction formed the basis ofthe study. A number of machine learning algorithms available within Waikato Environment for Knowledge Discovery had been evaluated and the most successful was chosen. The predictive model was subsequently validated in a different population of 931 patients (validation cohort), hospitalized during 2012. The best prediction results were achieved using Alternating Decision Tree classifier, which was able to predict in-hospital mortality with 89% accuracy, and preserved good performance on validation cohort with 87% accuracy. Alternating Decision Tree classifier identified a subset of 6 attributes most relevant to mortality prediction: systolic and diastolic blood pressure, heart rate, left ventricular ejection fraction, age, and troponin value. Data mining approach enabled the authors to develop a model capable of predicting the in-hospital outcome following percutaneous coronary intervention. The model showed excellent sensitivity and specificity during internal validation.

Development of Web tools to predict axillary lymph node metastasis and pathological response to neoadjuvant chemotherapy in breast cancer patients.

Nomograms are a standard computational tool to predict the likelihood of an outcome using multiple available patient features. We have developed a more powerful data mining methodology, to predict axillary lymph node (AxLN) metastasis and response to neoadjuvant chemotherapy (NAC) in primary breast cancer patients. We developed websites to use these tools. The tools calculate the probability of AxLN metastasis (AxLN model) and pathological complete response to NAC (NAC model). As a calculation algorithm, we employed a decision tree-based prediction model known as the alternative decision tree (ADTree), which is an analog development of if-then type decision trees. An ensemble technique was used to combine multiple ADTree predictions, resulting in higher generalization abilities and robustness against missing values. The AxLN model was developed with training datasets (n=148) and test datasets (n=143), and validated using an independent cohort (n=174), yielding an area under the receiver operating characteristic curve (AUC) of 0.768. The NAC model was developed and validated with n=150 and n=173 datasets from a randomized controlled trial, yielding an AUC of 0.787. AxLN and NAC models require users to input up to 17 and 16 variables, respectively. These include pathological features, including human epidermal growth factor receptor 2 (HER2) status and imaging findings. Each input variable has an option of "unknown," to facilitate prediction for cases with missing values. The websites developed facilitate the use of these tools, and serve as a database for accumulating new datasets.

Alternating decision tree algorithm for assessing protein interaction reliability

This paper presents a machine learning approach for assessing the reliability of protein–protein interactions in a high-throughput dataset. We use an alternating decision tree algorithm to distinguish true interacting protein pairs from noisy high-throughput data using various biological attributes of interacting proteins. The alternating decision tree algorithm is used both for identifying discriminating biological features that could be used for assessing protein interaction reliability and for constructing a classifier to identify true positive interacting pairs. Experimental results show that the proposed approach has a good performance in distinguishing true interacting protein pairs from noisy protein–protein interaction data. Moreover, our alternating decision tree classifier supplemented with domain knowledge may be helpful to understand the biological conditions in connection with interacting protein pairs.

Spatial Distribution of Perfusion Abnormality in Acute MCA Occlusion is Associated with Likelihood of Later Recanalization

The aim of this study is to investigate whether different spatial perfusion-deficit patterns, which indicate differing compensatory mechanisms, can be recognized and used to predict recanalization success of intravenous fibrinolytic therapy in acute stroke patients. Twenty-seven acute stroke data sets acquired within 6 hours from symptom onset including diffusion- (DWI) and perfusion-weighted magnetic resonance (MR) imaging (PWI) were analyzed and dichotomized regarding recanalization outcome using time-of-flight follow-up data sets. The DWI data sets were used for calculation of apparent diffusion coefficient (ADC) maps and subsequent infarct core segmentation. A patient-individual three-dimensional (3D) shell model was generated based on the segmentation and used for spatial analysis of the ADC as well as cerebral blood volume (CBV), cerebral blood flow, time to peak (TTP), and mean transit time (MTT) parameters derived from PWI. Skewness, kurtosis, area under the curve, and slope were calculated for each parameter curve and used for classification (recanalized/nonrecanalized) using a LogitBoost Alternating Decision Tree (LAD Tree). The LAD tree optimization revealed that only ADC skewness, CBV kurtosis, and MTT kurtosis are required for best possible prediction of recanalization success with a precision of 85%. Our results suggest that the propensity for macrovascular recanalization after intravenous fibrinolytic therapy depends not only on clot properties but also on distal microvascular bed perfusion. The 3D approach for characterization of perfusion parameters seems promising for further research.

Melanoma risk prediction models.

The lack of effective therapy for advanced stages of melanoma emphasizes the importance of preventive measures and screenings of population at risk. Identifying individuals at high risk should allow targeted screenings and follow-up involving those who would benefit most. The aim of this study was to identify most significant factors for melanoma prediction in our population and to create prognostic models for identification and differentiation of individuals at risk. This case-control study included 697 participants (341 patients and 356 controls) that underwent extensive interview and skin examination in order to check risk factors for melanoma. Pairwise univariate statistical comparison was used for the coarse selection of the most significant risk factors. These factors were fed into logistic regression (LR) and alternating decision trees (ADT) prognostic models that were assessed for their usefulness in identification of patients at risk to develop melanoma. Validation of the LR model was done by Hosmer and Lemeshow test, whereas the ADT was validated by 10-fold cross-validation. The achieved sensitivity, specificity, accuracy and AUC for both models were calculated. The melanoma risk score (MRS) based on the outcome of the LR model was presented. The LR model showed that the following risk factors were associated with melanoma: sunbeds (OR = 4.018; 95% CI 1.724-9.366 for those that sometimes used sunbeds), solar damage of the skin (OR = 8.274; 95% CI 2.661-25.730 for those with severe solar damage), hair color (OR = 3.222; 95% CI 1.984-5.231 for light brown/blond hair), the number of common naevi (over 100 naevi had OR = 3.57; 95% CI 1.427-8.931), the number of dysplastic naevi (from 1 to 10 dysplastic naevi OR was 2.672; 95% CI 1.572-4.540; for more than 10 naevi OR was 6.487; 95%; CI 1.993-21.119), Fitzpatricks phototype and the presence of congenital naevi. Red hair, phototype I and large congenital naevi were only present in melanoma patients and thus were strongly associated with melanoma. The percentage of correctly classified subjects in the LR model was 74.9%, sensitivity 71%, specificity 78.7% and AUC 0.805. For the ADT percentage of correctly classified instances was 71.9%, sensitivity 71.9%, specificity 79.4% and AUC 0.808. Application of different models for risk assessment and prediction of melanoma should provide efficient and standardized tool in the hands of clinicians. The presented models offer effective discrimination of individuals at high risk, transparent decision making and real-time implementation suitable for clinical practice. A continuous melanoma database growth would provide for further adjustments and enhancements in model accuracy as well as offering a possibility for successful application of more advanced data mining algorithms.

Prediction Of Allogeneic Hematopoietic Stem Cell Transplantation (allo-HSCT) Related Mortality in Acute Leukemia: Generation Of a Machine Learning-Based Model Using The Data Set of The Acute Leukemia Working Party (ALWP) Of The EBMT

▪ BackgroundAllo-HSCT has been shown to increase survival and improve cure in acute leukemia (AL). However, this procedure is accompanied by high rates of morbidity and mortality. Several risk scores based on conventional statistical methods may aid decision regarding whom and how to perform allo-HSCT, but these methods carry inherent limitations, which may lead to sub-optimal candidate selection. Machine learning (ML) is a field in computer science stemming from artificial intelligence and is part of the data mining approach for data analysis. ML algorithms are commonly applied in technological and commercial settings. They allow for coping with complex data scenarios and thus may be suitable for outcome prediction in allo-HSCT. With this background, and using a ML prediction method- the alternating decision tree (ADT) algorithm, we developed an interpretable model for overall mortality (OM) and treatment-related mortality (TRM) at day +100 after allo-HSCT in AL. Patients and Methods28,995 adult allo-HSCT recipients from the registry of the ALWP of EBMT were analyzed. Twenty two variables were available including year of transplant (range, 2000-2011), diagnosis (Acute Myeloid Leukemia and Acute Lymphoblastic Leukemia), disease status, Karnofsky performance status, conditioning regimen (myeloablative or reduced- intensity conditioning), graft type (peripheral blood, bone marrow or cord blood), donor and recipient HLA compatibility, CMV serostatus, GVHD prophylaxis regimens, etc. Per study definitions, the primary outcomes to be predicted were OM and TRM at day +100 days after allo-HSCT. The complete dataset was split into 3 sets: training set (n=11,600), testing set (n=8,688) and validation set (n=8,707). The ADT prediction model was tested and optimized according to the first 2 subgroups and validated on the last one. Output from the ADT model, included variables selection with assigned scores in a tree-based structure and area under the ROC curve (AUC), a measure for model discrimination ResultsEach of the ADT models selected 12 out 22 variables for prediction of OM and TRM at day +100. Ten variables were mutual for both prediction models, although different weights were assigned. These included: age, diagnosis, disease status, Karnofsky performance status (all at time of transplant), donor-recipient HLA-matching, number of transplants in each center per year, year of transplant, conditioning regimen and the donor's and patient's CMV serostatus. Variables selected exclusively by the OM prediction model were graft type and donor-patient CMV serostatus match, whereas the TRM model selected time from diagnosis to transplant and donor-recipient sex match. Applying the models on the validation set yielded AUCs of 0.701 (95% confidence interval [CI] 0.691-0.710) for OM prediction and 0.67 (95% [CI] 0.66-0.68) for TRM. The ADT prediction models assigned scores correlating with patient outcome. Patients in each of the validation sets were grouped according to their score range and the prediction success. A Higher score was correlated with higher rate of the measured outcome in both models (figure 1 and figure 2). [Display omitted] [Display omitted] ConclusionsWe present two new models, based on the ADT ML algorithm, for prediction of OM and TRM at day +100 after allo-HSCT. The models are robust as they rely on a high number of samples and a large validation set. As shown in the figures, higher scores correlated with a poorer outcome, reaching more than 50% mortality for a score range of 5.76-7 in the OM prediction model. The AUC performance measure was better for OM than TRM, possibly due to a higher event rate in former, making it easier to predict. Improving the predictive ability will probably necessitate evaluation of more variables, as the limitation of the maximal predictive performance is most likely in the information gained from the variables and not from the sample size or algorithm used. This is currently under progress, especially combination with other risk scores linked to comorbidities.In summary, our models can aid candidate selection for allo-HSCT, by providing a measurable score that correlates with transplant success. Disclosures:Schmid:Novartis: Honoraria, Research Funding, travel grant Other; Roche: travel grant, travel grant Other; MSD: Honoraria.