A real-world demonstration of machine learning generalizability in the detection of intracranial hemorrhage on head computerized tomography

AbstractMachine learning (ML) holds great promise in transforming healthcare. While published studies have shown the utility of ML models in interpreting medical imaging examinations, these are often evaluated under laboratory settings. The importance of real world evaluation is best illustrated by case studies that have documented successes and failures in the translation of these models into clinical environments. A key prerequisite for the clinical adoption of these technologies is demonstrating generalizable ML model performance under real world circumstances. The purpose of this study was to demonstrate that ML model generalizability is achievable in medical imaging with the detection of intracranial hemorrhage (ICH) on non-contrast computed tomography (CT) scans serving as the use case. An ML model was trained using 21,784 scans from the RSNA Intracranial Hemorrhage CT dataset while generalizability was evaluated using an external validation dataset obtained from our busy trauma and neurosurgical center. This real world external validation dataset consisted of every unenhanced head CT scan (n = 5965) performed in our emergency department in 2019 without exclusion. The model demonstrated an AUC of 98.4%, sensitivity of 98.8%, and specificity of 98.0%, on the test dataset. On external validation, the model demonstrated an AUC of 95.4%, sensitivity of 91.3%, and specificity of 94.1%. Evaluating the ML model using a real world external validation dataset that is temporally and geographically distinct from the training dataset indicates that ML generalizability is achievable in medical imaging applications.

Download Full-text

Discovery of Highly Polymorphic Organic Materials: A New Machine Learning Approach

10.26434/chemrxiv.9524219 ◽

2019 ◽

Author(s):

Zied Hosni ◽

Annalisa Riccardi ◽

Stephanie Yerdelen ◽

Alan R. G. Martin ◽

Deborah Bowering ◽

...

Keyword(s):

Machine Learning ◽

Structure Prediction ◽

External Validation ◽

New Drugs ◽

Training Dataset ◽

Validation Dataset ◽

Machine Learning Classification ◽

Novel Approach ◽

Physical Form ◽

Machine Learning Approach

<div><div><p>Polymorphism is the capacity of a molecule to adopt different conformations or molecular packing arrangements in the solid state. This is a key property to control during pharmaceutical manufacturing because it can impact a range of properties including stability and solubility. In this study, a novel approach based on machine learning classification methods is used to predict the likelihood for an organic compound to crystallise in multiple forms. A training dataset of drug-like molecules was curated from the Cambridge Structural Database (CSD) and filtered according to entries in the Drug Bank database. The number of separate forms in the CSD for each molecule was recorded. A metaclassifier was trained using this dataset to predict the expected number of crystalline forms from the compound descriptors. This approach was used to estimate the number of crystallographic forms for an external validation dataset. These results suggest this novel methodology can be used to predict the extent of polymorphism of new drugs or not-yet experimentally screened molecules. This promising method complements expensive ab initio methods for crystal structure prediction and as integral to experimental physical form screening, may identify systems that with unexplored potential.</p> </div> </div>

Download Full-text

Discovery of Highly Polymorphic Organic Materials: A New Machine Learning Approach

10.26434/chemrxiv.9524219.v1 ◽

2019 ◽

Author(s):

Zied Hosni ◽

Annalisa Riccardi ◽

Stephanie Yerdelen ◽

Alan R. G. Martin ◽

Deborah Bowering ◽

...

Keyword(s):

Machine Learning ◽

Structure Prediction ◽

External Validation ◽

New Drugs ◽

Training Dataset ◽

Validation Dataset ◽

Machine Learning Classification ◽

Novel Approach ◽

Physical Form ◽

Machine Learning Approach

Download Full-text

Predicting Future Occurrence of Acute Hypotensive Episodes Using Noninvasive and Invasive Features

Military Medicine ◽

10.1093/milmed/usaa418 ◽

2021 ◽

Vol 186 (Supplement_1) ◽

pp. 445-451

Author(s):

Yifei Sun ◽

Navid Rashedi ◽

Vikrant Vaze ◽

Parikshit Shah ◽

Ryan Halter ◽

...

Keyword(s):

Machine Learning ◽

Support Vector Machine ◽

Real World ◽

Short Term Memory ◽

Model Performance ◽

Learning Technologies ◽

Machine Learning Algorithms ◽

Support Vector ◽

K Nearest Neighbor ◽

Continuous Map

ABSTRACT Introduction Early prediction of the acute hypotensive episode (AHE) in critically ill patients has the potential to improve outcomes. In this study, we apply different machine learning algorithms to the MIMIC III Physionet dataset, containing more than 60,000 real-world intensive care unit records, to test commonly used machine learning technologies and compare their performances. Materials and Methods Five classification methods including K-nearest neighbor, logistic regression, support vector machine, random forest, and a deep learning method called long short-term memory are applied to predict an AHE 30 minutes in advance. An analysis comparing model performance when including versus excluding invasive features was conducted. To further study the pattern of the underlying mean arterial pressure (MAP), we apply a regression method to predict the continuous MAP values using linear regression over the next 60 minutes. Results Support vector machine yields the best performance in terms of recall (84%). Including the invasive features in the classification improves the performance significantly with both recall and precision increasing by more than 20 percentage points. We were able to predict the MAP with a root mean square error (a frequently used measure of the differences between the predicted values and the observed values) of 10 mmHg 60 minutes in the future. After converting continuous MAP predictions into AHE binary predictions, we achieve a 91% recall and 68% precision. In addition to predicting AHE, the MAP predictions provide clinically useful information regarding the timing and severity of the AHE occurrence. Conclusion We were able to predict AHE with precision and recall above 80% 30 minutes in advance with the large real-world dataset. The prediction of regression model can provide a more fine-grained, interpretable signal to practitioners. Model performance is improved by the inclusion of invasive features in predicting AHE, when compared to predicting the AHE based on only the available, restricted set of noninvasive technologies. This demonstrates the importance of exploring more noninvasive technologies for AHE prediction.

Download Full-text

Effectiveness of transfer learning for enhancing tumor classification with a convolutional neural network on frozen sections

Scientific Reports ◽

10.1038/s41598-020-78129-0 ◽

2020 ◽

Vol 10 (1) ◽

Author(s):

Young-Gon Kim ◽

Sungchul Kim ◽

Cristina Eunbee Cho ◽

In Hye Song ◽

Hee Jin Lee ◽

...

Keyword(s):

Neural Network ◽

Deep Learning ◽

Convolutional Neural Network ◽

Transfer Learning ◽

Frozen Section ◽

Medical Center ◽

External Validation ◽

Model Performance ◽

Classification Model ◽

Training Dataset

AbstractFast and accurate confirmation of metastasis on the frozen tissue section of intraoperative sentinel lymph node biopsy is an essential tool for critical surgical decisions. However, accurate diagnosis by pathologists is difficult within the time limitations. Training a robust and accurate deep learning model is also difficult owing to the limited number of frozen datasets with high quality labels. To overcome these issues, we validated the effectiveness of transfer learning from CAMELYON16 to improve performance of the convolutional neural network (CNN)-based classification model on our frozen dataset (N = 297) from Asan Medical Center (AMC). Among the 297 whole slide images (WSIs), 157 and 40 WSIs were used to train deep learning models with different dataset ratios at 2, 4, 8, 20, 40, and 100%. The remaining, i.e., 100 WSIs, were used to validate model performance in terms of patch- and slide-level classification. An additional 228 WSIs from Seoul National University Bundang Hospital (SNUBH) were used as an external validation. Three initial weights, i.e., scratch-based (random initialization), ImageNet-based, and CAMELYON16-based models were used to validate their effectiveness in external validation. In the patch-level classification results on the AMC dataset, CAMELYON16-based models trained with a small dataset (up to 40%, i.e., 62 WSIs) showed a significantly higher area under the curve (AUC) of 0.929 than those of the scratch- and ImageNet-based models at 0.897 and 0.919, respectively, while CAMELYON16-based and ImageNet-based models trained with 100% of the training dataset showed comparable AUCs at 0.944 and 0.943, respectively. For the external validation, CAMELYON16-based models showed higher AUCs than those of the scratch- and ImageNet-based models. Model performance for slide feasibility of the transfer learning to enhance model performance was validated in the case of frozen section datasets with limited numbers.

Download Full-text

Deep Learning Classification of Canine Behavior Using a Single Collar-Mounted Accelerometer: Real-World Validation

Animals ◽

10.3390/ani11061549 ◽

2021 ◽

Vol 11 (6) ◽

pp. 1549

Author(s):

Robert D. Chambers ◽

Nathanael C. Yoder ◽

Aletha B. Carson ◽

Christian Junge ◽

David E. Allen ◽

...

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Real World ◽

Learning Algorithm ◽

Drinking Behavior ◽

True Positive Rate ◽

Training Dataset ◽

Activity Levels ◽

Accelerometer Data ◽

Activity Monitors

Collar-mounted canine activity monitors can use accelerometer data to estimate dog activity levels, step counts, and distance traveled. With recent advances in machine learning and embedded computing, much more nuanced and accurate behavior classification has become possible, giving these affordable consumer devices the potential to improve the efficiency and effectiveness of pet healthcare. Here, we describe a novel deep learning algorithm that classifies dog behavior at sub-second resolution using commercial pet activity monitors. We built machine learning training databases from more than 5000 videos of more than 2500 dogs and ran the algorithms in production on more than 11 million days of device data. We then surveyed project participants representing 10,550 dogs, which provided 163,110 event responses to validate real-world detection of eating and drinking behavior. The resultant algorithm displayed a sensitivity and specificity for detecting drinking behavior (0.949 and 0.999, respectively) and eating behavior (0.988, 0.983). We also demonstrated detection of licking (0.772, 0.990), petting (0.305, 0.991), rubbing (0.729, 0.996), scratching (0.870, 0.997), and sniffing (0.610, 0.968). We show that the devices’ position on the collar had no measurable impact on performance. In production, users reported a true positive rate of 95.3% for eating (among 1514 users), and of 94.9% for drinking (among 1491 users). The study demonstrates the accurate detection of important health-related canine behaviors using a collar-mounted accelerometer. We trained and validated our algorithms on a large and realistic training dataset, and we assessed and confirmed accuracy in production via user validation.

Download Full-text

Systematic literature review of machine learning methods used in the analysis of real-world data for patient-provider decision making

BMC Medical Informatics and Decision Making ◽

10.1186/s12911-021-01403-2 ◽

2021 ◽

Vol 21 (1) ◽

Author(s):

Alan Brnabic ◽

Lisa M. Hess

Keyword(s):

Machine Learning ◽

Decision Making ◽

Literature Review ◽

Systematic Literature Review ◽

Real World ◽

Learning Algorithms ◽

External Validation ◽

Machine Learning Algorithms ◽

Learning Methods ◽

Machine Learning Methods

Abstract Background Machine learning is a broad term encompassing a number of methods that allow the investigator to learn from the data. These methods may permit large real-world databases to be more rapidly translated to applications to inform patient-provider decision making. Methods This systematic literature review was conducted to identify published observational research of employed machine learning to inform decision making at the patient-provider level. The search strategy was implemented and studies meeting eligibility criteria were evaluated by two independent reviewers. Relevant data related to study design, statistical methods and strengths and limitations were identified; study quality was assessed using a modified version of the Luo checklist. Results A total of 34 publications from January 2014 to September 2020 were identified and evaluated for this review. There were diverse methods, statistical packages and approaches used across identified studies. The most common methods included decision tree and random forest approaches. Most studies applied internal validation but only two conducted external validation. Most studies utilized one algorithm, and only eight studies applied multiple machine learning algorithms to the data. Seven items on the Luo checklist failed to be met by more than 50% of published studies. Conclusions A wide variety of approaches, algorithms, statistical software, and validation strategies were employed in the application of machine learning methods to inform patient-provider decision making. There is a need to ensure that multiple machine learning approaches are used, the model selection strategy is clearly defined, and both internal and external validation are necessary to be sure that decisions for patient care are being made with the highest quality evidence. Future work should routinely employ ensemble methods incorporating multiple machine learning algorithms.

Download Full-text

Proposing a machine-learning based method to predict stillbirth before and during delivery and ranking the features: nationwide retrospective cross-sectional study

BMC Pregnancy and Childbirth ◽

10.1186/s12884-021-03658-z ◽

2021 ◽

Vol 21 (1) ◽

Author(s):

Toktam Khatibi ◽

Elham Hanifi ◽

Mohammad Mehdi Sepehri ◽

Leila Allahqoli

Keyword(s):

Machine Learning ◽

External Validation ◽

Fetal Loss ◽

Null Distribution ◽

Training Dataset ◽

Gradient Boosting ◽

Support Vector ◽

Cross Sectional ◽

Boosting Method ◽

Demographic Features

Abstract Background Stillbirth is defined as fetal loss in pregnancy beyond 28 weeks by WHO. In this study, a machine-learning based method is proposed to predict stillbirth from livebirth and discriminate stillbirth before and during delivery and rank the features. Method A two-step stack ensemble classifier is proposed for classifying the instances into stillbirth and livebirth at the first step and then, classifying stillbirth before delivery from stillbirth during the labor at the second step. The proposed SE has two consecutive layers including the same classifiers. The base classifiers in each layer are decision tree, Gradient boosting classifier, logistics regression, random forest and support vector machines which are trained independently and aggregated based on Vote boosting method. Moreover, a new feature ranking method is proposed in this study based on mean decrease accuracy, Gini Index and model coefficients to find high-ranked features. Results IMAN registry dataset is used in this study considering all births at or beyond 28th gestational week from 2016/04/01 to 2017/01/01 including 1,415,623 live birth and 5502 stillbirth cases. A combination of maternal demographic features, clinical history, fetal properties, delivery descriptors, environmental features, healthcare service provider descriptors and socio-demographic features are considered. The experimental results show that our proposed SE outperforms the compared classifiers with the average accuracy of 90%, sensitivity of 91%, specificity of 88%. The discrimination of the proposed SE is assessed and the average AUC of ±95%, CI of 90.51% ±1.08 and 90% ±1.12 is obtained on training dataset for model development and test dataset for external validation, respectively. The proposed SE is calibrated using isotopic nonparametric calibration method with the score of 0.07. The process is repeated 10,000 times and AUC of SE classifiers using random different training datasets as null distribution. The obtained p-value to assess the specificity of the proposed SE is 0.0126 which shows the significance of the proposed SE. Conclusions Gestational age and fetal height are two most important features for discriminating livebirth from stillbirth. Moreover, hospital, province, delivery main cause, perinatal abnormality, miscarriage number and maternal age are the most important features for classifying stillbirth before and during delivery.

Download Full-text

KDClassifier: Urinary Proteomic Spectra Analysis Based on Machine Learning for Classification of Kidney Diseases

10.1101/2020.12.01.20242198 ◽

2020 ◽

Author(s):

Wanjun Zhao ◽

Yong Zhang ◽

Xinming Li ◽

Yonghong Mao ◽

Changwei Wu ◽

...

Keyword(s):

Machine Learning ◽

Mass Spectrum ◽

Kidney Disease ◽

Kidney Diseases ◽

Training Dataset ◽

Validation Dataset ◽

Support Vector ◽

Urinary Proteomics ◽

Diagnosis Model

AbstractBackgroundBy extracting the spectrum features from urinary proteomics based on an advanced mass spectrometer and machine learning algorithms, more accurate reporting results can be achieved for disease classification. We attempted to establish a novel diagnosis model of kidney diseases by combining machine learning with an extreme gradient boosting (XGBoost) algorithm with complete mass spectrum information from the urinary proteomics.MethodsWe enrolled 134 patients (including those with IgA nephropathy, membranous nephropathy, and diabetic kidney disease) and 68 healthy participants as a control, and for training and validation of the diagnostic model, applied a total of 610,102 mass spectra from their urinary proteomics produced using high-resolution mass spectrometry. We divided the mass spectrum data into a training dataset (80%) and a validation dataset (20%). The training dataset was directly used to create a diagnosis model using XGBoost, random forest (RF), a support vector machine (SVM), and artificial neural networks (ANNs). The diagnostic accuracy was evaluated using a confusion matrix. We also constructed the receiver operating-characteristic, Lorenz, and gain curves to evaluate the diagnosis model.ResultsCompared with RF, the SVM, and ANNs, the modified XGBoost model, called a Kidney Disease Classifier (KDClassifier), showed the best performance. The accuracy of the diagnostic XGBoost model was 96.03% (CI = 95.17%-96.77%; Kapa = 0.943; McNemar’s Test, P value = 0.00027). The area under the curve of the XGBoost model was 0.952 (CI = 0.9307-0.9733). The Kolmogorov-Smirnov (KS) value of the Lorenz curve was 0.8514. The Lorenz and gain curves showed the strong robustness of the developed model.ConclusionsThis study presents the first XGBoost diagnosis model, i.e., the KDClassifier, combined with complete mass spectrum information from the urinary proteomics for distinguishing different kidney diseases. KDClassifier achieves a high accuracy and robustness, providing a potential tool for the classification of all types of kidney diseases.

Download Full-text

Machine Learning Algorithms to Predict Recurrence within 10 Years after Breast Cancer Surgery: A Prospective Cohort Study

Cancers ◽

10.3390/cancers12123817 ◽

2020 ◽

Vol 12 (12) ◽

pp. 3817

Author(s):

Shi-Jer Lou ◽

Ming-Feng Hou ◽

Hong-Tai Chang ◽

Chong-Chi Chiu ◽

Hao-Hsien Lee ◽

...

Keyword(s):

Breast Cancer ◽

Machine Learning ◽

Learning Algorithms ◽

External Validation ◽

Model Development ◽

Cancer Surgery ◽

Machine Learning Algorithms ◽

Breast Cancer Surgery ◽

Training Dataset

No studies have discussed machine learning algorithms to predict recurrence within 10 years after breast cancer surgery. This study purposed to compare the accuracy of forecasting models to predict recurrence within 10 years after breast cancer surgery and to identify significant predictors of recurrence. Registry data for breast cancer surgery patients were allocated to a training dataset (n = 798) for model development, a testing dataset (n = 171) for internal validation, and a validating dataset (n = 171) for external validation. Global sensitivity analysis was then performed to evaluate the significance of the selected predictors. Demographic characteristics, clinical characteristics, quality of care, and preoperative quality of life were significantly associated with recurrence within 10 years after breast cancer surgery (p < 0.05). Artificial neural networks had the highest prediction performance indices. Additionally, the surgeon volume was the best predictor of recurrence within 10 years after breast cancer surgery, followed by hospital volume and tumor stage. Accurate recurrence within 10 years prediction by machine learning algorithms may improve precision in managing patients after breast cancer surgery and improve understanding of risk factors for recurrence within 10 years after breast cancer surgery.

Download Full-text

Predicting Inpatient Length of Stay After Brain Tumor Surgery: Developing Machine Learning Ensembles to Improve Predictive Performance

Neurosurgery ◽

10.1093/neuros/nyy343 ◽

2018 ◽

Vol 85 (3) ◽

pp. 384-393 ◽

Cited By ~ 12

Author(s):

Whitney E Muhlestein ◽

Dallin S Akagi ◽

Jason M Davies ◽

Lola B Chambless

Keyword(s):

Machine Learning ◽

Brain Tumor ◽

Length Of Stay ◽

Patient Outcomes ◽

External Validation ◽

Training Dataset ◽

Ensemble Model ◽

Improvement Program ◽

Improve Patient ◽

Inpatient Length Of Stay

Abstract BACKGROUND Current outcomes prediction tools are largely based on and limited by regression methods. Utilization of machine learning (ML) methods that can handle multiple diverse inputs could strengthen predictive abilities and improve patient outcomes. Inpatient length of stay (LOS) is one such outcome that serves as a surrogate for patient disease severity and resource utilization. OBJECTIVE To develop a novel method to systematically rank, select, and combine ML algorithms to build a model that predicts LOS following craniotomy for brain tumor. METHODS A training dataset of 41 222 patients who underwent craniotomy for brain tumor was created from the National Inpatient Sample. Twenty-nine ML algorithms were trained on 26 preoperative variables to predict LOS. Trained algorithms were ranked by calculating the root mean square logarithmic error (RMSLE) and top performing algorithms combined to form an ensemble. The ensemble was externally validated using a dataset of 4592 patients from the National Surgical Quality Improvement Program. Additional analyses identified variables that most strongly influence the ensemble model predictions. RESULTS The ensemble model predicted LOS with RMSLE of .555 (95% confidence interval, .553-.557) on internal validation and .631 on external validation. Nonelective surgery, preoperative pneumonia, sodium abnormality, or weight loss, and non-White race were the strongest predictors of increased LOS. CONCLUSION An ML ensemble model predicts LOS with good performance on internal and external validation, and yields clinical insights that may potentially improve patient outcomes. This systematic ML method can be applied to a broad range of clinical problems to improve patient care.

Download Full-text