Final Classification Research Articles

Parallel ensemble of a randomization-based online sequential neural network for classification problems using a frequency criterion

Randomization-based neural networks have gained wide acceptance in the scientific community owing to the simplicity of their algorithm and generalization capabilities. Random vector functional link (RVFL) networks and their variants are a class of randomization-based neural networks. RVFL networks have shown promising results in classification, regression, and clustering problems. For real-world applications, learning algorithms that can train with new samples over previous results are necessary because of to the constant generation of problems related to large-scale datasets. Various online sequential algorithms, commonly involving an initial learning phase followed by a sequential learning phase, have been proposed to address this issue. This paper presents a training algorithm based on multiple online sequential random vector functional link (OS-RVFL) networks for large-scale databases using a shared memory architecture. The training dataset is distributed among p OS-RVFL networks, which are trained in parallel using p threads. Subsequently, the test dataset samples are classified using each trained OS-RVFL network. Finally, a frequency criterion is applied to the results obtained from each OS-RVFL network to determine the final classification. Additionally, an equation was derived to reasonably predict the total training time of the proposed algorithm based on the learning time in the initial phase and the time scaling factor compared to the sequential learning phase. The results demonstrate a drastic reduction in training time because of data distribution and an improvement in accuracy because of the adoption of the frequency criterion.

Morphological classification of neurons based on Sugeno fuzzy integration and multi-classifier fusion

In order to extract more important morphological features of neuron images and achieve accurate classification of the neuron type, a method is proposed that uses Sugeno fuzzy integral integration of three optimized deep learning models, namely AlexNet, VGG11_bn, and ResNet-50. Firstly, using the pre-trained model of AlexNet and the output layer is fine-tuned to improve the model’s performance. Secondly, in the VGG11_bn network, Global Average Pooling (GAP) is adopted to replace the traditional fully connected layer to reduce the number of parameters. Additionally, the generalization ability of the model is improved by transfer learning. Thirdly, the SE(squeeze and excitation) module is added to the ResNet-50 variant ResNeXt-50 to adjust the channel weight and capture the key information of the input data. The GELU activation function is used to better fit the data distribution. Finally, Sugeno fuzzy integral is used to fuse the output of each model to get the final classification result. The experimental results showed that on the Img_raw, Img_resample and Img_XYalign dataset, the accuracy of 4-category classification reached 98.04%, 91.75% and 93.13%, respectively, and the accuracy of 12-category classification reached 97.82%, 85.68% and 87.60%, respectively. The proposed method has good classification performance in the morphological classification of neurons.

Arrhythmia detection in inter-patient ECG signals using entropy rate features and RR intervals with CNN architecture

The classification of inter-patient ECG data for arrhythmia detection using electrocardiogram (ECG) signals presents a significant challenge. Despite the recent surge in deep learning approaches, there remains a noticeable gap in the performance of inter-patient ECG classification. In this study, we introduce an innovative approach for ECG classification in arrhythmia detection by employing a 1D convolutional neural network (CNN) to leverage both morphological and temporal characteristics of cardiac cycles. Through the utilization of 1D-CNN layers, we automatically capture the morphological attributes of ECG data, allowing us to represent the shape of the ECG waveform around the R peaks. Additionally, we incorporate four RR interval features to provide temporal context, and we explore the potential application of entropy rate as a feature extraction technique for ECG signal classification. Consequently, the classification layers benefit from the combination of both temporal and learned features, leading to the achievement of the final arrhythmia classification. We validate our approach using the MIT-BIH arrhythmia dataset, employing both intra-patient and inter-patient paradigms for model training and testing. The model’s generalization ability is assessed by evaluating it on the INCART dataset. The model attains average accuracy rates of 99.13% and 99.17% for 2-fold and 5-fold cross-validation, respectively, in intra-patient classification with five classes. In inter-patient classification with three and five classes, the model achieves average accuracies of 98.73% and 97.91%, respectively. For the INCART dataset, the model achieves an average accuracy of 98.20% for three classes. The experimental outcomes demonstrate the superiority of the proposed model compared to state-of-the-art models in recognizing arrhythmias. Thus, the proposed model exhibits enhanced generalization and the potential to serve as an effective solution for recognizing arrhythmias in real-world datasets characterized by class imbalances in practical applications.

Building a framework for fake news detection in the health domain.

Disinformation in the medical field is a growing problem that carries a significant risk. Therefore, it is crucial to detect and combat it effectively. In this article, we provide three elements to aid in this fight: 1) a new framework that collects health-related articles from verification entities and facilitates their check-worthiness and fact-checking annotation at the sentence level; 2) a corpus generated using this framework, composed of 10335 sentences annotated in these two concepts and grouped into 327 articles, which we call KEANE (faKe nEws At seNtence lEvel); and 3) a new model for verifying fake news that combines specific identifiers of the medical domain with triplets subject-predicate-object, using Transformers and feedforward neural networks at the sentence level. This model predicts the fact-checking of sentences and evaluates the veracity of the entire article. After training this model on our corpus, we achieved remarkable results in the binary classification of sentences (check-worthiness F1: 0.749, fact-checking F1: 0.698) and in the final classification of complete articles (F1: 0.703). We also tested its performance against another public dataset and found that it performed better than most systems evaluated on that dataset. Moreover, the corpus we provide differs from other existing corpora in its duality of sentence-article annotation, which can provide an additional level of justification of the prediction of truth or untruth made by the model.

ML-Based Edge Node for Monitoring Peoples' Frailty Status.

The development of contactless methods to assess the degree of personal hygiene in elderly people is crucial for detecting frailty and providing early intervention to prevent complete loss of autonomy, cognitive impairment, and hospitalisation. The unobtrusive nature of the technology is essential in the context of maintaining good quality of life. The use of cameras and edge computing with sensors provides a way of monitoring subjects without interrupting their normal routines, and has the advantages of local data processing and improved privacy. This work describes the development an intelligent system that takes the RGB frames of a video as input to classify the occurrence of brushing teeth, washing hands, and fixing hair. No action activity is considered. The RGB frames are first processed by two Mediapipe algorithms to extract body keypoints related to the pose and hands, which represent the features to be classified. The optimal feature extractor results from the most complex Mediapipe pose estimator combined with the most complex hand keypoint regressor, which achieves the best performance even when operating at one frame per second. The final classifier is a Light Gradient Boosting Machine classifier that achieves more than 94% weighted F1-score under conditions of one frame per second and observation times of seven seconds or more. When the observation window is enlarged to ten seconds, the F1-scores for each class oscillate between 94.66% and 96.35%.

Automatic diagnosis of epileptic seizures using entropy-based features and multimodel deep learning approaches

Epilepsy is one of the most common brain diseases, characterised by repeated seizures that occur on a regular basis. During a seizure, a patient's muscles flex uncontrollably, causing a loss of mobility and balance, which can be harmful or even fatal. Developing an automatic approach for warning patients of oncoming seizures necessitates substantial research. Analyzing the electroencephalogram (EEG) output from the human brain's scalp region can help predict seizures. EEG data were analyzed to extract time domain features such as Hurst exponent (Hur), Tsallis entropy (TsEn), enhanced permutation entropy (impe), and amplitude-aware permutation entropy (AAPE). In order to automatically diagnose epileptic seizure in children from normal children, this study conducted two sessions. In the first session, the extracted features from the EEG dataset were classified using three machine learning (ML)-based models, including support vector machine (SVM), K nearest neighbor (KNN), or decision tree (DT), and in the second session, the dataset was classified using three deep learning (DL)-based recurrent neural network (RNN) classifiers in The EEG dataset was obtained from the Neurology Clinic of the Ibn Rushd Training Hospital. In this regard, extensive explanations and research from the time domain and entropy characteristics demonstrate that employing GRU, LSTM, and BiLSTM RNN deep learning classifiers on the All−time−entropy fusion feature improves the final classification results.

Operational classification of cutaneous squamous cell carcinomas based on unsupervised clustering of real cases by experts.

There is currently no staging system for cutaneous squamous cell carcinoma (cSCC) that is adapted to decision-making and universally used. Experts have unconscious ability to simplify the heterogeneity of clinical situations into a few relevant groups to drive their therapeutic decisions. Therefore, we have used unsupervised clustering of real cases by experts to generate an operational classification of cSCCs, an approach that was successful for basal cell carcinomas. To generate a consensual and operational classification of cSCCs. Unsupervised independent clustering of 248 cases of cSCCs considered difficult-to-treat. Eighteen international experts from different specialties classified these cases into what they considered homogeneous clusters useful for management, each with freedom regarding clustering criteria. Convergences and divergences between clustering were analysed using a similarity matrix, the K-mean approach and the average silhouette method. Mathematical modelling was used to look for the best consensual clustering. The operability of the derived classification was validated on 23 new practitioners. Despite the high heterogeneity of the clinical cases, a mathematical consensus was observed. It was best represented by a partition into five clusters, which appeared a posteriori to describe different clinical scenarios. Applicability of this classification was shown by a good concordance (94%) in the allocation of cases between the new practitioners and the 18 experts. An additional group of easy-to-treat cSCC was included, resulting in a six-group final classification: easy-to-treat/complex to treat due to tumour and/or patient characteristics/multiple/locally advanced/regional disease/visceral metastases. Given the methodology based on the convergence of unguided intuitive clustering of cases by experts, this new classification is relevant for clinical practice. It does not compete with staging systems, but they may complement each other, whether the objective is to select the best therapeutic approach in tumour boards or to design homogeneous groups for trials.

Complications and adverse events in lymphadenectomy of the inguinal area: worldwide expert consensus.

Inguinal lymph node dissection plays an important role in the management of melanoma, penile and vulval cancer. Inguinal lymph node dissection is associated with various intraoperative and postoperative complications with significant heterogeneity in classification and reporting. This lack of standardization challenges efforts to study and report inguinal lymph node dissection outcomes. The aim of this study was to devise a system to standardize the classification and reporting of inguinal lymph node dissection perioperative complications by creating a worldwide collaborative, the complications and adverse events in lymphadenectomy of the inguinal area (CALI) group. A modified 3-round Delphi consensus approach surveyed a worldwide group of experts in inguinal lymph node dissection for melanoma, penile and vulval cancer. The group of experts included general surgeons, urologists and oncologists (gynaecological and surgical). The survey assessed expert agreement on inguinal lymph node dissection perioperative complications. Panel interrater agreement and consistency were assessed as the overall percentage agreement and Cronbach's α. Forty-seven experienced consultants were enrolled: 26 (55.3%) urologists, 11 (23.4%) surgical oncologists, 6 (12.8%) general surgeons and 4 (8.5%) gynaecology oncologists. Based on their expertise, 31 (66%), 10 (21.3%) and 22 (46.8%) of the participants treat penile cancer, vulval cancer and melanoma using inguinal lymph node dissection respectively; 89.4% (42 of 47) agreed with the definitions and inclusion as part of the inguinal lymph node dissection intraoperative complication group, while 93.6% (44 of 47) agreed that postoperative complications should be subclassified into five macrocategories. Unanimous agreement (100%, 37 of 37) was achieved with the final standardized classification system for reporting inguinal lymph node dissection complications in melanoma, vulval cancer and penile cancer. The complications and adverse events in lymphadenectomy of the inguinal area classification system has been developed as a tool to standardize the assessment and reporting of complications during inguinal lymph node dissection for the treatment of melanoma, vulval and penile cancer.

Multi-Scale and Multi-Factor ViT Attention Model for Classification and Detection of Pest and Disease in Agriculture

Agriculture has a crucial impact on the economic, ecological, and social development of the world. More rapid and precise prevention and control work, especially for accurate classification and detection, is required due to the increasing severity of agricultural pests and diseases. However, the results of the image classification and detection are unsatisfactory because of the limitation of image data volume acquisition and the wide range of influencing factors of pests and diseases. In order to solve these problems, the vision transformer (ViT) model is improved, and a multi-scale and multi-factor ViT attention model (SFA-ViT) is proposed in this paper. Data augmentation considering multiple influencing factors is implemented in SFA-ViT to mitigate the impact of insufficient experimental data. Meanwhile, SFA-ViT optimizes the ViT model from a multi-scale perspective, and encourages the model to understand more features, from fine-grained to coarse-grained, during the classification task. Further, the detection model based on the self-attention mechanism of the multi-scale ViT is constructed to achieve the accurate localization of the pest and disease. Finally, experimental validation of the model, based on the IP102 and Plant Village dataset, is carried out. The results indicate that the various components of SFA-ViT effectively enhance the final classification and detection outcomes, and our model outperforms the current models significantly.

Natural Language Processing (NLP) Accurately Identifies LTCF Exposure from Clinical Notes: A Proof-of-Principle Study

Background: Residence or recent stay in a long-term care facility (LTCF) is one of the most important risk factors for multidrug-resistant organism (MDRO) carriage and infection, making reliable identification of LTCF-exposed inpatients a critical priority for infection control day-to-day practice and research. However, because most hospital electronic health records (EHRs) do not include a dedicated field for documenting LTCF exposure, absent manual review of patient charts, identifying LTCF-exposed inpatients is challenging. We aimed to develop an automated, natural language processing (NLP)-based classifier for identifying LTCF exposure from clinical notes. Methods: We randomly sampled 1020 adult admissions from 2016-2021 across the 12-hospital University of Maryland Medical System and manually reviewed each admission’s history & physical (H&P) note for mention of LTCF exposure (Figure 1). After H&P pre-processing, we calculated feature representations for documents based on term frequencies and visually explored between-group (LTCF-exposed vs. LTCF-unexposed) feature differences. To predict LTCF status from the H&P notes, we trained and tuned a LASSO regression-based classifier on 70% of the data with 3-fold cross-validation and 1:1 up-sampling to address class imbalance. The final classifier was evaluated on the 30% held-out sample (not up-sampled), with calculation of the C-statistic (area-under-the-curve, AUC) with bootstrapped 95% confidence intervals, and construction of receiver-operating-characteristic and variable importance plots (R Version 4.3.2). Results: 7% (n=76 cases) of H&P notes documented LTCF exposure. In our visual analysis, the H&P words and phrases that were over-represented among LTCF patients had high face validity (Figure 2). The final LASSO-regression-based classifier achieved a C-statistic of 0.89 (95% CI: 0.80–0.98) on the held-out data for identifying LTCF exposure from the H&P notes (Figure 3). The most important model predictors (i.e., words) for distinguishing LTCF-exposed from LTCF-unexposed patients are reflected in Figure 4. The most important predictor-words of LTCF-exposure were “rehab,” “place,” “status,” “egd,” and “dementia.” Conclusion: In this multi-center study, even a simple NLP classifier demonstrated very strong discrimination for identifying LTCF exposure status from H&P notes, which could substantially reduce the manual review time required to identify LTCF-exposed inpatients. If automated in the electronic health record, it could also inform real-time MDRO screening decisions. Future research is planned to build more sophisticated classifiers using machine learning best practices, to build classifiers for additional MDRO risk factors, and to externally validate NLP classifiers on notes from an external healthcare system.

Fully Unsupervised Deepfake Video Detection Via Enhanced Contrastive Learning.

Nowadays, Deepfake videos are widely spread over the Internet, which severely impairs the public trustworthiness and social security. Although more and more reliable detectors have recently sprung up for resisting against that new-emerging tampering technique, some challengeable issues still need to be addressed, such that most of Deepfake video detectors under the framework of the supervised mechanism require a large scale of samples with accurate labels for training. When the amount of the training samples with the true labels are not enough or the training data are maliciously poisoned by adversaries, the supervised classifier is probably not reliable for detection. To tackle that tough issue, it is proposed to design a fully unsupervised Deepfake detector. In particular, in the whole procedure of training or testing, we have no idea of any information about the true labels of samples. First, we novelly design a pseudo-label generator for labeling the training samples, where the traditional hand-crafted features are used to characterize both types of samples. Second, the training samples with the pseudo-labels are fed into the proposed enhanced contrastive learner, in which the discriminative features are further extracted and continually refined by iteration on the guidance of the contrastive loss. Last, relying on the inter-frame correlation, we complete the final binary classification between real and fake videos. A large scale of experimental results empirically verify the effectiveness of our proposed unsupervised Deepfake detector on the benchmark datasets including FF++, Celeb-DF, DFD, DFDC, and UADFV. Furthermore, our proposed well-performed detector is superior to the current unsupervised method, and comparable to the baseline supervised methods. More importantly, when facing the problem of the labeled data poisoned by malicious adversaries or insufficient data for training, our proposed unsupervised Deepfake detector performs its powerful superiority.

A Novel Part Refinement Tandem Transformer for Human-Object Interaction Detection.

Human-object interaction (HOI) detection identifies a "set of interactions" in an image involving the recognition of interacting instances and the classification of interaction categories. The complexity and variety of image content make this task challenging. Recently, the Transformer has been applied in computer vision and received attention in the HOI detection task. Therefore, this paper proposes a novel Part Refinement Tandem Transformer (PRTT) for HOI detection. Unlike the previous Transformer-based HOI method, PRTT utilizes multiple decoders to split and process rich elements of HOI prediction and introduces a new part state feature extraction (PSFE) module to help improve the final interaction category classification. We adopt a novel prior feature integrated cross-attention (PFIC) to utilize the fine-grained partial state semantic and appearance feature output obtained by the PSFE module to guide queries. We validate our method on two public datasets, V-COCO and HICO-DET. Compared to state-of-the-art models, the performance of detecting human-object interaction is significantly improved by the PRTT.

Learning With Incremental Instances and Features.

In many real-world applications, data may dynamically expand over time in both volume and feature dimensions. Besides, they are often collected in batches (also called blocks). We refer this kind of data whose volume and features increase in blocks as blocky trapezoidal data streams. Current works either assume that the feature space of data streams is fixed or stipulate that the algorithm receives only one instance at a time, and none of them can effectively handle the blocky trapezoidal data streams. In this article, we propose a novel algorithm to learn a classification model from blocky trapezoidal data streams, called learning with incremental instances and features (IIF). We attempt to design highly dynamic model update strategies that can learn from increasing training data with an expanding feature space. Specifically, we first divide the data streams obtained on each round and construct the corresponding classifiers for these different divided parts. Then, to realize the effective interaction of information between each classifier, we utilize a single global loss function to capture their relationship. Finally, we use the idea of ensemble to achieve the final classification model. Furthermore, to make this method more applicable, we directly transform it into the kernel method. Both theoretical analysis and empirical analysis validate the effectiveness of our algorithm.

Achieving Better Category Separability for Hyperspectral Image Classification: A Spatial-Spectral Approach.

The task of hyperspectral image (HSI) classification has attracted extensive attention. The rich spectral information in HSIs not only provides more detailed information but also brings a lot of redundant information. Redundant information makes spectral curves of different categories have similar trends, which leads to poor category separability. In this article, we achieve better category separability from the perspective of increasing the difference between categories and reducing the variation within category, thus improving the classification accuracy. Specifically, we propose the template spectrum-based processing module from spectral perspective, which can effectively expose the unique characteristics of different categories and reduce the difficulty of model mining key features. Second, we design an adaptive dual attention network from spatial perspective, where the target pixel can adaptively aggregate high-level features by evaluating the confidence of effective information in different receptive fields. Compared with the single adjacency scheme, the adaptive dual attention mechanism makes the ability of target pixel to combine spatial information to reduce variation more stable. Finally, we designed a dispersion loss from the classifier's perspective. By supervising the learnable parameters of the final classification layer, the loss makes the category standard eigenvectors learned by the model more dispersed, which improves the category separability and reduces the rate of misclassification. Experiments on three common datasets show that our proposed method is superior to the comparison method.

Klasifikasi Jenis Jamur Edible Menggunakan Convolutional Neural Network: Studi Kasus pada Jamur Tiram, Enoki, dan Truffle

This study uses a Convolutional Neural Network (CNN) to develop a mushroom type classification model that can differentiate between truffles, enoki and oyster mushrooms very accurately. The dataset consisting of mushroom images is collected from various sources and processed through data augmentation techniques such as rotation, zoom, flip, and shifting to improve model generalization. For final classification, the CNN model used consists of several convolution and pooling layers, followed by a dense layer. The study results show that this model has a validation accuracy of 79.31% after twenty training epochs. Although there were some misclassifications between oyster and enoki mushrooms, data augmentation techniques were shown to improve model performance. This research shows that CNN can be used effectively to classify various types of fungi. This also shows that CNN has many potential applications in the agricultural and food industries. Increasing datasets, studying more complex model architectures, and using advanced data augmentation methods such as Generative Adversarial Networks (GANs) are some suggestions for further research.

E-DNet: An End-to-End Dual-Branch Network for Driver Steering Intention Detection

An advanced driving assistant system (ADAS) is critical for improving traffic efficiency and ensuring driving safety. By anticipating the driver’s steering intentions in advance, the system can alert the driver in time to avoid a vehicle collision. This paper proposes a novel end-to-end dual-branch network (EDNet) that utilizes both in-cabin and out-of-cabin data. In this study, we designed an in-cabin driver intent feature extractor based on 3D residual networks and atrous convolution, which is applicable to video data and is capable of capturing a larger range of driver behavior. In order to capture the long-term dependency of temporal data, we designed the depthwise-separable max-pooling (DSMax) module and combined it with a convolutional LSTM to obtain the road environment feature extractor outside the cabin. In addition, to effectively fuse different features inside and outside the cockpit, we designed and propose the dynamic combined-feature attention fusion (D-CAF) module. EDNet employs a freeze-training method, which enables the creation of a lightweight model while simultaneously enhancing the final classification accuracy. Extensive experiments on the Brain4Cars dataset and the Zenodo dataset show that the proposed EDNet was able to recognize the driver’s steering intention up to 3 s in advance. It outperformed the existing state of the art in most driving scenarios.

Enhancing the interpretability of sentiment analysis with signatures

Sentiment analysis is a significant research area within the field of natural language processing, with applications in various social contexts. Recent advances in related research have led to notable improvements in accuracy, yet existing text sentiment analysis models continue to face challenges in terms of interpretability. This paper proposes a text sentiment classification method integrating BERT and BiLstm, comprising three stages: training, signature generation, and testing. Our primary contribution lies in the signature generation and testing stages. In the training stage, BERT is employed to generate word embeddings, while BiLstm is utilized for learning to construct the text sentiment classification model. In the signature generation stage, the training and validation sets are passed through the aforementioned classification model, the feature vectors are extracted following their passage through the fully-connected layer, and the feature vectors are utilized to generate the signature generation stage. The signature is then stored in the database. In the Testing Stage, the aforementioned operation is repeated on the test set in order to obtain the signatures. The signatures are then matched in the constructed signature database in order to obtain the final classification results. The matching degree is subsequently used as a threshold in order to be employed for subsequent evaluation of the results. The method enhances the interpretability of the classification task by matching through signatures. Two datasets, SST-2 and IMDB, were subjected to experimentation. Four quantitative metrics were employed for the evaluation of the results: accuracy, recall, precision, and F1-score. With a threshold of 0.99, the models optimal accuracy, recall, precision, and F1-score results reached 0.896, 1, 0.896, 0.945 and 0.847, 0.999, 0.848, 0.917, respectively.

Craniocervical Morphometry in Pomeranians-Part II: Associations with Chiari-like Malformation and Syringomyelia.

The aim of Part II of this two-part study is to describe and analyze the association of various aspects and measurements related to the morphometry of the skull and craniocervical region to CM/SM status of Pomeranians, by means of computed tomography (CT) and magnetic resonance imaging (MRI). Prospectively, Pomeranians were included that underwent both CT and MRI studies of the head and cervicothoracic vertebral column. For those cases where qualitative classifications differed between observers, the experienced observer re-evaluated the studies and decided on a final classification that was used for further analysis. For quantitative measurements, the means of the observers' measurements were used for analysis. Among statistically significant differences in measurements, we found that dogs with SM had a significantly shorter clivus length based on both MRI (p = 0.01) and CT measurements (p = 0.01), and a significantly smaller caudal cranial fossa area based on both MRI (p = 0.02) and CT measurements (p = 0.02). Significant morphometrical differences were identified between dogs with or without CM/SM. The findings in this study add to those already described in other breeds and provide further insight into factors that may play a role in the pathogenesis of CM/SM in Pomeranians.

Measuring quality of life in Europe: A new fuzzy multicriteria approach

Quality of life indicators include measures of both objective and subjective well-being to broaden the traditional income-focused view towards a multidimensional view of welfare. In this paper, we develop a single reliable composite index employing fuzzy mathematical modelling that considers the effect of uncertainty and imprecision related to the specific nature of the phenomenon being measured. Additionally, we suggest a procedure for incorporating this new way of assigning priorities through multicriteria techniques. To this end, we propose a new multicriteria approach that introduces the recent notion of picture fuzzy sets to deal with uncertainty and imprecision in the assessment of objective and subjective dimensions of quality-of-life performance. To verify the effectiveness of the whole approach and the credibility of the results, we show its application for monitoring the quality of life in the European Union. Our results show that fuzzy numbers can be a useful tool when combined with multicriteria techniques in the prioritization phase of the criteria and that they increase the quality and reliability of the overall aggregate indicator. Furthermore, through statistical measures we found that in a picture environment the results are consistent regardless of the multi-criteria approach used to provide a final classification.

Development and Validation of a Deep Learning Classifier Using Chest Radiographs to Predict Extubation Success in Patients Undergoing Invasive Mechanical Ventilation.

The decision to extubate patients on invasive mechanical ventilation is critical; however, clinician performance in identifying patients to liberate from the ventilator is poor. Machine Learning-based predictors using tabular data have been developed; however, these fail to capture the wide spectrum of data available. Here, we develop and validate a deep learning-based model using routinely collected chest X-rays to predict the outcome of attempted extubation. We included 2288 serial patients admitted to the Medical ICU at an urban academic medical center, who underwent invasive mechanical ventilation, with at least one intubated CXR, and a documented extubation attempt. The last CXR before extubation for each patient was taken and split 79/21 for training/testing sets, then transfer learning with k-fold cross-validation was used on a pre-trained ResNet50 deep learning architecture. The top three models were ensembled to form a final classifier. The Grad-CAM technique was used to visualize image regions driving predictions. The model achieved an AUC of 0.66, AUPRC of 0.94, sensitivity of 0.62, and specificity of 0.60. The model performance was improved compared to the Rapid Shallow Breathing Index (AUC 0.61) and the only identified previous study in this domain (AUC 0.55), but significant room for improvement and experimentation remains.