Explainable Machine Learning Research Articles

Explainable Machine Learning Models for Real-Time Threat Detection in Cybersecurity

In the rapidly evolving landscape of cybersecurity, traditional machine learning models often operate as "black boxes," providing high accuracy but lacking transparency in decision-making. This lack of explainability poses challenges for trust and accountability, especially in critical areas like threat detection and incident response. Explainable machine learning models aim to address this by making the model's predictions more understandable and interpretable to users. This research integrates explainable machine learning models for real-time threat detection in cybersecurity. Data from multiple sources, including network traffic, system logs, and user behavior, undergo preprocessing such as cleaning, feature extraction, and normalization. The processed data is passed through various machine learning models, including traditional approaches like SVM and decision trees, as well as deep learning models like CNN and RNN. Explainability techniques such as LIME, SHAP, and attention mechanisms provide transparency, ensuring interpretable predictions. The explanations are delivered through a user interface that generates alerts, visualizations, and reports, facilitating effective threat assessment and incident response in decision support systems. This framework enhances model performance, trust, and reliability in complex cybersecurity scenarios.

Explainable machine learning methods to predict postpartum depression risk

Postpartum depression (PPD) is a type of depression that mothers have following childbirth due to hormonal changes, psychological transition to parenting, and exhaustion. This depression strikes either during/or in the first year following childbirth. It is also a frequently disregarded medical condition that must be treated right away as it might have major repercussions. Machine learning (ML) and artificial intelligence (AI) are tools that healthcare professionals can utilize to anticipate this condition more rapidly and correctly. Consequently, we have demonstrated how to use explainable artificial intelligence (XAI) methods and heterogeneous classifiers to predict postpartum depression in mothers who have recently given birth. The K-Nearest Neighbor (KNN) model and the customized stack model outperformed all other classifiers. KNN model obtained 97% accuracy, 98% recall, and 95% precision and the stack model obtained 97% accuracy, 100% recall, and 94% precision, respectively. A set of frameworks and resources known as explainable artificial intelligence (XAI) facilitates the comprehension and interpretation of predictions made by machine learning algorithms. Four distinct XAI techniques: ELI5, Shapley Additive Values (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Anchor – have been used to interpret the model predictions. Explainability, interpretability, accountability, and transparency are crucial parameters of XAI, ensuring that machine learning models provide understandable and trustworthy results to users and stakeholders. The goal of this interdisciplinary research is to develop an automated diagnosis framework with tools that can transform therapy for postpartum depression leading to suicide attempts and empower medical professionals to offer mothers individualized, high-quality care.

Mortality Prediction Modeling for Patients with Breast Cancer Based on Explainable Machine Learning

Background/Objectives: Breast cancer is the most common cancer in women worldwide, requiring strategic efforts to reduce its mortality. This study aimed to develop a predictive classification model for breast cancer mortality using real-world data, including various clinical features. Methods: A total of 11,286 patients with breast cancer from the National Cancer Center were included in this study. The mortality rate of the total sample was approximately 6.2%. Propensity score matching was used to reduce bias. Several machine learning models, including extreme gradient boosting, were applied to 31 clinical features. To enhance model interpretability, we used the SHapley Additive exPlanations method. ML analyses were also performed on the samples, excluding patients who developed other cancers after breast cancer. Results: Among the ML models, the XGB model exhibited the highest discriminatory power, with an area under the curve of 0.8722 and a specificity of 0.9472. Key predictors of the mortality classification model included occurrence in other organs, age at diagnosis, N stage, T stage, curative radiation treatment, and Ki-67(%). Even after excluding patients who developed other cancers after breast cancer, the XGB model remained the best-performing, with an AUC of 0.8518 and a specificity of 0.9766. Additionally, the top predictors from SHAP were similar to the results for the overall sample. Conclusions: Our models provided excellent predictions of breast cancer mortality using real-world data from South Korea. Explainable artificial intelligence, such as SHAP, validated the clinical applicability and interpretability of these models.

Abstract 4134947: Explainable Machine Learning to Predict Heart Failure Outcomes Using Exercise and Clinical Data

Background: Heart failure (HF) hospitalizations are associated with disease burden and cardiovascular mortality. International guidelines highlight cardiopulmonary testing (CPET) for risk stratification and prognosis in HF. Identifying individuals at risk could direct interventions to prevent admissions. This study uses explainable machine learning (XML) tools to identify CPET and clinical variables predictive of HF hospitalizations. Hypothesis: XML analysis of CPET parameters and clinical factors can yield accurate, patient-specific prediction of HF admission within one year. Methods: A retrospective single center review of 178 CPETs in patients referred for HF was completed. Clinical data within one year of CPET trained a boosted machine learning model (XGBoost). SHAP (Shapley Additive Explanations) explainability analysis emphasized key features for model predictions. Results: 44 (24.7%) patients with CPET were admitted for HF within one year. The model predicted this outcome with an area under the receiver operating characteristic curve (AUC) of 0.772. SHAP analysis (Figure 1) selected factors for decision making: pulmonary artery systolic pressure (PASP) via cardiac catheterization or echocardiogram; oxygen consumption at the ventilatory threshold (VO 2 at VT); O 2 pulse at rest; ventilatory efficiency (VE/VCO 2 slope); minute ventilation per oxygen consumption at maximal exercise (VE/VO 2 max) during CPET; left ventricle end diastolic volume by echocardiogram; and sodium-glucose cotransporter-2 inhibitor usage. Peak VO2, predicted VO2, predicted O2 pulse, and VE/VCO2 slope were significant, consistent with past studies (Table 1). Conclusion: Our XML analysis emphasizes PASP (partially related to high left ventricular filling pressures) and VO 2 (at ventilatory threshold) as predictive for HF hospitalization within one year of CPET testing. Studies to prospectively evaluate these highlighted factors are needed to continue exploring predictive capabilities to tailor interventions.

Abstract 4136721: Reducing the Use of Transesophageal Echocardiogram Prior to Cardioversion Using Machine Learning

Introduction/Background: Direct current cardioversion (DCC) is used to terminate atrial fibrillation and atrial flutter. Due to increased stroke risk post DCC, a transesophageal echocardiogram (TEE) is performed in selected patients to rule out left atrial appendage thrombus (LAAT) prior to DCC. The prevalence of LAAT is around 2.5% in the era of direct oral anticoagulants. This is a large proportion of patients with no LAAT on TEE and calls for strategies to improve patient selections for TEE before DCC. Research Questions/Hypothesis: In this study we aimed to determine if the use of machine learning could accurately predict the presence of LAAT among patients with nonvalvular atrial fibrillation and atrial flutter. Methods/Approach: We collected demographic, comorbid, and transthoracic echocardiographic data on 795 patients who underwent TEE prior to DCC for nonvalvular atrial fibrillation and atrial flutter. We developed and validated an explainable machine learning model using the eXtreme Gradient Boosting algorithm with 5x5 nested cross-validation. The algorithm used 37 variables to predict the probability of LAAT. Results/Data: The prevalence of LAAT in our cohort was 11.3%. Our model achieved an area under the curve (AUC) of 0.79 in predicting LAAT, figure 1A. In comparison, CHA2DS2-VASc score performed poorly in our cohort with an AUC of 0.48, figure 1B. The ranking of variables most strongly predictive of LAAT is shown in figure 1C, with left ventricle ejection fraction as the most instrumental. Patients were categorized into 10 groups based on the percentile of their predicted probability of having a thrombus. A lower percentile (e.g., 10%) indicates a lower probability of having a thrombus. By utilizing a cutoff point of 0.16, which includes 10.0% of the patients, we can exclude the presence of thrombus with 100% certainty. Conclusion: Machine learning, compared to traditional risk scores, offers enhanced predictive model for identifying LAAT and provides better guidance for TEE and anticoagulation therapy before DCC.

Explainable machine learning for early prediction of sepsis in traumatic brain injury: A discovery and validation study.

People with traumatic brain injury (TBI) are at high risk for infection and sepsis. The aim of the study was to develop and validate an explainable machine learning(ML) model based on clinical features for early prediction of the risk of sepsis in TBI patients. We enrolled all patients with TBI in the Medical Information Mart for Intensive Care IV database from 2008 to 2019. All patients were randomly divided into a training set (70%) and a test set (30%). The univariate and multivariate regression analyses were used for feature selection. Six ML methods were applied to develop the model. The predictive performance of different models were determined based on the area under the curve (AUC) and calibration curves in the test cohort. In addition, we selected the eICU Collaborative Research Database version 1.2 as the external validation dataset. Finally, we used the Shapley additive interpretation to account for the effects of features attributed to the model. Of the 1555 patients enrolled in the final cohort, 834 (53.6%) patients developed sepsis after TBI. Six variables were associated with concomitant sepsis and were used to develop ML models. Of the 6 models constructed, the Extreme Gradient Boosting (XGB) model achieved the best performance with an AUC of 0.807 and an accuracy of 74.5% in the internal validation cohort, and an AUC of 0.762 for the external validation. Feature importance analysis revealed that use mechanical ventilation, SAPSII score, use intravenous pressors, blood transfusion on admission, history of diabetes, and presence of post-stroke sequelae were the top six most influential features of the XGB model. As shown in the study, the ML model could be used to predict the occurrence of sepsis in patients with TBI in the intensive care unit.

Explainable machine learning model for predicting the risk of significant liver fibrosis in patients with diabetic retinopathy.

Diabetic retinopathy (DR), a prevalent complication in patients with type 2 diabetes, has attracted increasing attention. Recent studies have explored a plausible association between retinopathy and significant liver fibrosis. The aim of this investigation was to develop a sophisticated machine learning (ML) model, leveraging comprehensive clinical datasets, to forecast the likelihood of significant liver fibrosis in patients with retinopathy and to interpret the ML model by applying the SHapley Additive exPlanations (SHAP) method. This inquiry was based on data from the National Health and Nutrition Examination Survey 2005-2008 cohort. Utilizing the Fibrosis-4 index (FIB-4), liver fibrosis was stratified across a spectrum of grades (F0-F4). The severity of retinopathy was determined using retinal imaging and segmented into four discrete gradations. A ten-fold cross-validation approach was used to gauge the propensity towards liver fibrosis. Eight ML methodologies were used: Extreme Gradient Boosting, Random Forest, multilayer perceptron, Support Vector Machines, Logistic Regression (LR), Plain Bayes, Decision Tree, and k-nearest neighbors. The efficacy of these models was gauged using metrics, such as the area under the curve (AUC). The SHAP method was deployed to unravel the intricacies of feature importance and explicate the inner workings of the ML model. The analysis included 5,364 participants, of whom 2,116 (39.45%) exhibited notable liver fibrosis. Following random allocation, 3,754 individuals were assigned to the training set and 1,610 were allocated to the validation cohort. Nine variables were curated for integration into the ML model. Among the eight ML models scrutinized, the LR model attained zenith in both AUC (0.867, 95% CI: 0.855-0.878) and F1 score (0.749, 95% CI: 0.732-0.767). In internal validation, this model sustained its superiority, with an AUC of 0.850 and an F1 score of 0.736, surpassing all other ML models. The SHAP methodology unveils the foremost factors through importance ranking. Sophisticated ML models were crafted using clinical data to discern the propensity for significant liver fibrosis in patients with retinopathy and to intervene early. Improved early detection of liver fibrosis risk in retinopathy patients enhances clinical intervention outcomes.

EPCO-36. INTEGRATIVE DEEP-LEARNING AND PHARMACOGENOMIC PROFILING OF MALIGNANT GLIOMAS USING PATIENT-DERIVED TUMOR CELLS

Abstract Malignant glioma stands as a formidable cancer with limited therapeutic options due to many factors including pronounced genetic heterogeneity, the blood-brain barrier, among others. Precision medicine, tailoring treatment based on a patient’s unique molecular profiles, is a fitting approach for addressing intra- and inter-tumor heterogeneity. Previously, we have shown the advantage of using patient-derived tumor cells (PDCs) to evaluate drug efficacy across various cancer types. Unlike cancer cell lines, which significantly differ from primary tumors, PDCs capture the molecular features of primary tumors, showing predictive clinical responses in a retrospective study. Building on these encouraging results, we have expanded PDC-based drug screening to specifically target malignant gliomas. We have established an initial cohort of over 500 malignant glioma PDCs (~ 370 patients) and screened more than 100 drugs. We have complemented our cohort with molecular profiles, including genomic and transcriptomic data to infer the potential pharmacogenomic associations of malignant gliomas. We constructed a multi-modal deep-learning model that seamlessly integrates molecular features and drug structure to predict the drug sensitivity. By employing an explainable machine learning method, we retrospectively analyzed the model to investigate the potential molecular biomarkers and their contribution to drug sensitivity. Collectively, our study demonstrates the efficacy of using patient-derived tumor cells (PDCs) combined with deep-learning models to predict drug sensitivity in malignant gliomas. This integrative approach can potentially refine personalized treatment strategies and improve clinical outcomes for glioma patients.

Explainable machine learning model for predicting molten steel temperature in the LF refining process

Explainable machine learning model for predicting molten steel temperature in the LF refining process

Advancing buffet onset prediction: a deep learning approach with enhanced interpretability for aerodynamic engineering

The interaction between the shock wave and boundary layer of transonic wings can trigger periodic self-excited oscillations, resulting in transonic buffet. Buffet severely restricts the flight envelope of civil aircraft and is directly related to their aerodynamic performance and safety. Developing efficient and reliable techniques for buffet onset prediction is crucial for the advancement of civil aircraft. In this study, utilizing a comprehensive database of supercritical airfoils generated through numerical simulations, a convolutional neural network (CNN) model is firstly developed to perform buffet classification based on the flow fields. After that, employing explainable machine learning techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM), random forest algorithms, and statistical analysis, the research investigates the correlations between supervised CNN features and key physical characteristics related with the separation region, shock wave, leading edge suction peak, and post-shock loading. Finally, physical buffet onset metric is established with good generalization and accuracy, providing valuable guidance for engineering design in civil aircraft.

Explainable machine learning models for predicting the acute toxicity of pesticides to sheepshead minnow (Cyprinodon variegatus).

A quantitative structure-activity relationship (QSAR) study was conducted on 313 pesticides to predict their acute toxicity to Sheepshead minnow (Cyprinodon variegatus) by using DRAGON descriptors. Essentials accounting for a reliable model were all considered carefully, giving full consideration to the OECD (Organization for Economic Co-operation and Development) principles for QSAR acceptability in regulation during the model construction and assessment process. Nine variables were selected through the forward stepwise regression method and used as inputs to construct both linear and nonlinear models. The obtained models were validated internally and externally. Generally, machine learning-based methods, namely support vector machine (SVM), random forest (RF), and projection pursuit regression (PPR), perform better than the multiple linear regression (MLR) model. The statistical results (R2 = 0.682-0.933, Q2LOO = 0.604-0.659, Q2F1 = 0.740-0.796, CCC = 0.861-0.882) of the developed models show that they are robust, reliable, reproducible, accurate and predictive. Comparatively, the RF model performs best, giving predictive correlation coefficient Q2 of 0.814, root mean squared error (RMSE) of 0.658 and mean absolute error (MAE) of 0.534 for the test set, respectively. The RF model (as well as SVM and PPR models) was visualized and explained by using the SHapley Additive explanation (SHAP) analysis to enhance its transparency and credibility. In addition, the applicability domain (AD) range of the RF model was characterized by the Williams plot and the tree manifold approximation and projection (TMAP) technology was utilized to illustrate similarity and diversity of the entire data space, to assist in the analysis of the outliers. Activity cliff detection was investigated by using Arithmetic Residuals in K-groups Analysis (ARKA) descriptors. It was found that none of the pesticides was identified as an activity cliff in the training set or a potential prediction cliff in the test set. Therefore, the RF model fulfills each OECD principle in regulation for QSAR models. The research in this work will aid in the in silico QSAR prediction of the acute toxicity to Sheepshead minnow (Cyprinodon variegatus) for untested and new toxic pesticides and can also be extended to other studies.

Observation impact explanation in atmospheric state estimation using hierarchical message-passing graph neural networks**This paper is significantly revised from an earlier version [] presented at the Explainable Machine Learning for Sciences Workshop (XAI4Sci) in conjunction with the 38th AAAI Conference on Artificial Intelligence (AAAI 2024) held in Vancouver, Canada, in February 2024.

Abstract The impact of meteorological observations on weather forecasting varies with the sensor type, location, time, and other environmental factors. Thus, the quantitative analysis of observation impacts is crucial for the effective and efficient development of weather forecasting systems. However, existing impact analysis methods are dependent on specific forecast systems, because system-specific adjoint models are used and the sensitivity of the observation to the forecast is measured. This study investigates the impact of observations on atmospheric state estimation in weather forecasting systems by developing a novel graph neural network (GNN) model specialized for analyzing the heterogeneous relations between observations and atmospheric states. The observation impact can then be assessed by applying explainable methods to the proposed GNN model, which is independent of forecasting systems. Further, we develop a novel application called ‘CloudNine,’ a system that provides impact analysis for individual observations with visualization. Our GNN model comprises hierarchical message-passing modules that separately analyze spatial correlations between observations at close locations and atmospheric states at close locations and then examine correlations between observations and atmospheric states. To consider the different factors influencing these correlations, we utilized geo-coordinates and types of observations in the attention mechanism of the modules with their feature vectors. We then applied gradient-based explainability methods to quantify the significance of the different observations in the estimation. Evaluated using data from 11 satellites and land-based observations, the results highlight the effectiveness of the proposed model and the visualization of observation impacts, enhancing the understanding and optimization of observational data in weather forecasting.

Supplemental Material for Understanding the Psychological, Relational, Sociocultural, and Demographic Predictors of Loneliness Using Explainable Machine Learning

Supplemental Material for Understanding the Psychological, Relational, Sociocultural, and Demographic Predictors of Loneliness Using Explainable Machine Learning

Exploring a Data‐Driven Approach to Identify Regions of Change Associated With Future Climate Scenarios

AbstractA key consideration for evaluating climate projections is uncertainty in future radiative forcing scenarios. Although it is straightforward to monitor greenhouse gas concentrations and compare observations with specified climate scenarios, it remains less obvious how to detect and attribute regional pattern changes with plausible future mitigation scenarios. Here we introduce a machine learning approach for linking patterns of climate change with radiative forcing scenarios and use a feature attribution method to understand how these linkages are made. We train a neural network using output from the SPEAR Large Ensemble to classify whether temperature or precipitation maps are most likely to originate from one of several potential radiative forcing scenarios. Despite substantial atmospheric internal variability, the neural network learns to identify “fingerprint” patterns, including significant localized regions of change, that associate specific patterns of climate change with radiative forcing scenarios in each year of the simulations. We illustrate this using output from additional ensembles with sharp reductions in future greenhouse gases and highlight specific regions (in this example, the subpolar North Atlantic and Central Africa) that are critical for associating the new simulations with changes in radiative forcing scenarios. Overall, this framework suggests that explainable machine learning could provide one strategy for detecting a regional climate response to future mitigation efforts.

Enhancing and Explaining Energy Management in Smart Manufacturing Production via ML Within the Context of Industry 4.0

Objectives: This study aims to explore the integration of big data and machine learning to enhance energy efficiency in manufacturing, focusing on the opportunities presented by Industry 4.0 and cyber-physical production systems. Theoretical Framework: The research leverages supervised learning techniques to analyze and predict machine-specific energy consumption patterns, utilizing energy disaggregation as a foundational concept. Method: Various machine learning algorithms, including Lasso regression, Linear regression, and Decision Tree, will be employed to develop predictive models for energy usage. Additionally, Explainable Machine Learning (XML) techniques will be utilized to ensure interpretability and clarity in the prediction outcomes. Results and Discussion: The proposed framework aims to elucidate the relationship between equipment utility and energy consumption, providing comprehensible explanations that enhance decision-making processes within intelligent industrial environments. Research Implications: This work highlights the significant potential of XML in transforming machine learning applications in manufacturing, paving the way for improved energy efficiency and operational effectiveness. Originality/Value: The introduction of an innovative framework that combines machine learning with explainability in the context of energy consumption marks a valuable contribution to the fields of manufacturing and sustainable industry practices.

Data-driven explainable machine learning for personalized risk classification of myasthenic crisis

ObjectiveMyasthenic crisis (MC) is a critical progression of Myasthenia gravis (MG), requiring intensive care treatment and invasive therapies. Classifying patients at high-risk for MC facilitates treatment decisions such as changes in medication or the need for mechanical ventilation and helps prevent disease progression by decreasing treatment-induced stress on the patient. Here, we investigated whether it is possible to reliably classify MG patients into groups at low or high risk of MC based entirely on routine medical data using explainable machine learning (ML). MethodsIn this single-center pseudo-prospective cohort study, we investigated the precision of ML models trained with real-world routine clinical data to identify MG patients at risk for MC, and identified explainable distinctive features for the groups. 51 MG patients, including 13 MC, were used for model training based on real-world clinical data available from the hospital management system. Patients were classified to high or low risk for MC using Lasso regression or random forest ML models. ResultsThe mean cross-validated AUC classifying MG patients as high or low risk for MC based on simple or compound features derived from real-world clinical data showed a predictive accuracy of 68.8% for a regularized Lasso regression and 76.5% for a random forest model. Studying feature importance across 5100 model runs identified explainable features to distinguish MG patients at high or low risk for MC. Feature importance scores suggested that multimorbidity may play a role in risk classification. ConclusionThis study establishes feasibility and proof-of-concept for risk classification of MC based on real-world routine clinical data using ML with explainable features and variance control at the point of care. Future research on ML-based prediction of MC should include multi-center, multinational data collection, more in-depth data per patient, more patients, and an attention-based ML model to include free-text.

Explainable machine learning by SEE-Net: closing the gap between interpretable models and DNNs

Deep Neural Networks (DNNs) have achieved remarkable accuracy for numerous applications, yet their complexity often renders the explanation of predictions a challenging task. This complexity contrasts with easily interpretable statistical models, which, however, often suffer from lower accuracy. Our work suggests that this underperformance may stem more from inadequate training methods than from the inherent limitations of model structures. We hereby introduce the Synced Explanation-Enhanced Neural Network (SEE-Net), a novel architecture integrating a guiding DNN with a shallow neural network, functionally equivalent to a two-layer mixture of linear models. This shallow network is trained under the guidance of the DNN, effectively bridging the gap between the prediction power of deep learning and the need for explainable models. Experiments on image and tabular data demonstrate that SEE-Net can leverage the advantage of DNNs while providing an interpretable prediction framework. Critically, SEE-Net embodies a new paradigm in machine learning: it achieves high-level explainability with minimal compromise on prediction accuracy by training an almost “white-box” model under the co-supervision of a “black-box” model, which can be tailored for diverse applications.

Data augmented planning: A data-driven approach to measuring-understanding-optimizing green justice across 263 Chinese cities

Urban Green Spaces (UGS) are pivotal in fostering sustainable urban environments, with disparities in access to UGS raising concerns about 'green justice'. Existing research often focuses on the socioeconomic influences on green justice, overlooking the role of urban green planning and design. To address this, we evaluated green justice levels across 263 Chinese cities using extensive geospatial data and explainable machine learning techniques. Our findings highlight a significant relationship between the planar configuration of UGS and green justice. Notably, these associations are non-linear, showing threshold effects where certain UGS characteristics beyond specific values can adversely affect green justice. The study also identifies interaction effects among these spatial characteristics. Our results offer practical guidelines for urban planning to enhance green justice, especially in China's large cities. Recommendations include optimizing green space morphology and proximity, and reconfiguring the layout of various green spaces. This research provides valuable insights and strategies for promoting green justice, particularly in densely populated cities in developing nations like China.

Stiffness and hardness of thermally modified timber assessed with explainable machine learning

Thermal modification is an environmentally friendly practice to enhance the durability of wood. Despite the method's effectiveness in improving wood durability and dimensional stability, the mechanical properties of wood could be compromised following the heat treatment process. While most of the literature has focused on studying the impact of wood modification on stiffness or strength, less attention was given to characterization and nondestructive prediction of wood hardness following thermal treatment. Hardness, an important quality indicator of wood, has been a subject of limited research in the context of developing predictive models for assessing the hardness of thermally modified wood. The existing research in this area has either been scarce or unsuccessful, with poor predictive performance being a common issue. This research studies the impact of thermal modification on the longitudinal and transverse hardness of some common North American softwoods (Douglas-fir, Western hemlock, red cedar) and hardwoods (red alder, paper birch). It also aims to develop an explainable machine-learning (TreeNet gradient boosting machine) predictive model for hardness prediction using stress wave data combined with wood density and information about the type of wood species and heat treatment level. The impact of heat treatment on the stress wave velocity, density, and hardness was studied, and the dependency of the variables was assessed using correlation and clustering analysis. Machine learning modeling showed that the longitudinal and transverse hardness could be predicted by having a maximum R2 (test data) of 70.62 and 72.78, respectively, with a relative importance of the predictors in descending order as wood species > density > wave velocity >MOEdyn> heat treatment. Using the top 3 critical parameters yielded an R2 (test data) of ∼69 % for both models. The developed model could predict the hardness of various wood species treated at different temperatures, having only the type of wood species and its density followed by the stress wave velocity, which can be measured at an industrial scale.

Explainable AI for automated respiratory misalignment detection in PET/CT imaging

Purpose.Positron emission tomography (PET) image quality can be affected by artifacts emanating from PET, computed tomography (CT), or artifacts due to misalignment between PET and CT images. Automated detection of misalignment artifacts can be helpful both in data curation and in facilitating clinical workflow. This study aimed to develop an explainable machine learning approach to detect misalignment artifacts in PET/CT imaging.Approach.This study included 1216 PET/CT images. All images were visualized and images with respiratory misalignment artifact (RMA) detected. Using previously trained models, four organs including the lungs, liver, spleen, and heart were delineated on PET and CT images separately. Data were randomly split into cross-validation (80%) and test set (20%), then two segmentations performed on PET and CT images were compared and the comparison metrics used as predictors for a random forest framework in a 10-fold scheme on cross-validation data. The trained models were tested on 20% test set data. The model's performance was calculated in terms of specificity, sensitivity, F1-Score and area under the curve (AUC).Main results.Sensitivity, specificity, and AUC of 0.82, 0.85, and 0.91 were achieved in ten-fold data split. F1_score, sensitivity, specificity, and AUC of 84.5 vs 82.3, 83.9 vs 83.8, 87.7 vs 83.5, and 93.2 vs 90.1 were achieved for cross-validation vs test set, respectively. The liver and lung were the most important organs selected after feature selection.Significance.We developed an automated pipeline to segment four organs from PET and CT images separately and used the match between these segmentations to decide about the presence of misalignment artifact. This methodology may follow the same logic as a reader detecting misalignment through comparing the contours of organs on PET and CT images. The proposed method can be used to clean large datasets or integrated into a clinical scanner to indicate artifactual cases.