Deep learning for air pollutant forecasting: opportunities, challenges, and future directions

The growing availability of spatial data from remote sensing, laser scanning (LiDAR), and photogrammetric techniques stimulates the dynamic development of methods for the automatic detection and classification of topographic objects. In recent years, both classical machine learning (ML) algorithms and deep learning (DL) methods have found wide application in the analysis of large and complex data sets. Despite significant achievements, literature on the subject remains scattered, and a comprehensive review that systematically compares algorithm classes with respect to data modality, performance, and application context is still needed. The aim of this article is to provide a critical analysis of the current state of research on the use of ML and DL algorithms in the detection and classification of topographic objects. The theoretical foundations of selected methods, their applications to various data sources, and the accuracy and computational requirements reported in the literature are presented. Attention is paid to comparing classical ML algorithms (including SVM, RF, KNN) with modern deep architectures (CNN, U-Net, ResNet), with respect to different data types such as satellite imagery, aerial orthophotos, and LiDAR point clouds, indicating their effectiveness in the context of cartographic and elevation data. The article also discusses the main challenges related to data availability, model interpretability, and computational costs, and points to promising directions for further research. The summary of the results shows that DL methods are frequently reported to achieve several to over ten percentage points higher segmentation and classification accuracy than classical ML approaches, depending on data type and object complexity, particularly in the analysis of raster data and LiDAR point clouds. The conclusions emphasize the practical significance of these methods for spatial planning, infrastructure monitoring, and environmental management, as well as their potential in the automation of topographic analysis.

Comparing ECG-Lead Subsets for Heart Arrhythmias/ECG Patterns Classification: Convolutional Neural Networks and Random Forest

Deep learning methods have become popular among researchers in the field of fault detection. However, their performance depends on the availability of big datasets. To overcome this problem researchers started applying transfer learning to achieve good performance from small available datasets, by leveraging multiple prediction models over similar machines and working conditions. However, the influence of negative transfer limits their application. Negative transfer among prediction models increases when the environment and working conditions are changing continuously. To overcome the effect of negative transfer, we propose a novel deep transfer learning method, coined deep boosted transfer learning, for wind turbine gearbox fault detection that prevents negative transfer and only focuses on relevant information from the source machine. The proposed method is an instance-based deep transfer learning method that updates the weights of the source and the target machine training samples separately. The weights of different source training samples are gradually decreased to reduce the impact on the final model. The proposed method is verified by the Case Western Reserve University bearing and real field wind farm datasets. The results show that the proposed method ignores negative transfer and achieves higher accuracy compared to standard deep learning and deep transfer learning methods.

Assessing Cancer Risk from Mammograms: Deep Learning Is Superior to Conventional Risk Models.

The paper proposes a new method for classifying the LISS IV satellite images using deep learning method. Deep learning method is to automatically extract many features without any human intervention. The classification accuracy through deep learning is still improved by including object-based segmentation. The object-based deep feature learning method using CNN is used to accurately classify the remotely sensed images. The method is designed with the technique of extracting the deep features and using it for object-based classification. The proposed system extracts deep features using pre-defined filter values, thus increasing the overall performance of the process compared to randomly initialized filter values. The object-based classification method can preserve edge information in complex satellite images. To improve the classification accuracy and to reduce complexity, object-based deep learning technique is used. The proposed object-based deep learning approach is used to drastically increase the classification accuracy. Here, the remotely sensed images were used to classify the urban areas of Ahmadabad and Madurai cities. Experimental results show a better performance with the object-based classification.

Tomato production can be greatly reduced due to various diseases, such as bacterial spot, early blight, and leaf mold. Rapid recognition and timely treatment of diseases can minimize tomato production loss. Nowadays, a large number of researchers (including different institutes, laboratories, and universities) have developed and examined various traditional machine learning (ML) and deep learning (DL) algorithms for plant disease classification. However, through pass survey analysis, we found that there are no studies comparing the classification performance of ML and DL for the tomato disease classification problem. The performance and outcomes of different traditional ML and DL (a subset of ML) methods may vary depending on the datasets used and the tasks to be solved. This study generally aimed to identify the most suitable ML/DL models for the PlantVillage tomato dataset and the tomato disease classification problem. For machine learning algorithm implementation, we used different methods to extract disease features manually. In our study, we extracted a total of 52 texture features using local binary pattern (LBP) and gray level co-occurrence matrix (GLCM) methods and 105 color features using color moment and color histogram methods. Among all the feature extraction methods, the COLOR+GLCM method obtained the best result. By comparing the different methods, we found that the metrics (accuracy, precision, recall, F1 score) of the tested deep learning networks (AlexNet, VGG16, ResNet34, EfficientNet-b0, and MobileNetV2) were all better than those of the measured machine learning algorithms (support vector machine (SVM), k-nearest neighbor (kNN), and random forest (RF)). Furthermore, we found that, for our dataset and classification task, among the tested ML/DL algorithms, the ResNet34 network obtained the best results, with accuracy of 99.7%, precision of 99.6%, recall of 99.7%, and F1 score of 99.7%.

Cardiac perfusion PET is commonly used to assess ischemia and cardiovascular risk, which enables quantitative measurements of myocardial blood flow (MBF) through kinetic modeling. However, the estimation of kinetic parameters is challenging due to the noisy nature of short dynamic frames and limited sample data points. This work aimed to investigate the errors in MBF estimation in PET through a simulation study and to evaluate different parameter estimation approaches, including a deep learning (DL) method. Simulated studies were generated using digital phantoms based on cardiac segmentations from 55 clinical CT images. We employed the irreversible 2-tissue compartmental model and simulated dynamic 13N-ammonia PET scans under both rest and stress conditions (220 cases each). The simulations covered a rest K1 range of 0.6 to 1.2 and a stress K1 range of 1.2 to 3.6 (unit: mL/min/g) in the myocardium. A transformer-based DL model was trained on the simulated dataset to predict parametric images (PIMs) from noisy PET image frames and was validated using 5-fold cross-validation. We compared the DL method with the voxel-wise nonlinear least squares (NLS) fitting applied to the dynamic images, using either Gaussian filter (GF) smoothing (GF-NLS) or a dynamic nonlocal means (DNLM) algorithm for denoising (DNLM-NLS). Two patients with coronary CT angiography (CTA) and fractional flow reserve (FFR) were enrolled to test the feasibility of applying DL models on clinical PET data. The DL method showed clearer image structures with reduced noise compared to the traditional NLS-based methods. In terms of mean absolute relative error (MARE), as the rest K1 values increased from 0.6 to 1.2 mL/min/g, the overall bias in myocardium K1 estimates decreased from approximately 58% to 45% for the NLS-based methods while the DL method showed a reduction in MARE from 42% to 18%. For stress data, as the stress K1 decreased from 3.6 to 1.2mL/min/g, the MARE increased from 30% to 70% for the GF-NLS method. In contrast, both the DNLM-NLS (average: 42%) and the DL methods (average: 20%) demonstrated significantly smaller MARE changes as stress K1 varied. Regarding the regional mean bias (±standard deviation), the GF-NLS method had a bias of 6.30% (±8.35%) of rest K1, compared to 1.10% (±8.21%) for DNLM-NLS and 6.28% (±14.05%) for the DL method. For the stress K1, the GF-NLS showed a mean bias of 10.72% (±9.34%) compared to 1.69% (±8.82%) for DNLM-NLS and -10.55% (±9.81%) for the DL method. This study showed that an increase in the tracer uptake rate (K1) corresponded to improved accuracy and precision in MBF quantification, whereas lower tracer uptake resulted in higher noise in dynamic PET and poorer parameter estimates. Utilizing denoising techniques or DL approaches can mitigate noise-induced bias in PET parametric imaging.

Less than 10% of patients with cancer have detectable pathogenic germline alterations, which may be partially due to incomplete pathogenic variant detection. To evaluate if deep learning approaches identify more germline pathogenic variants in patients with cancer. A cross-sectional study of a standard germline detection method and a deep learning method in 2 convenience cohorts with prostate cancer and melanoma enrolled in the US and Europe between 2010 and 2017. The final date of clinical data collection was December 2017. Germline variant detection using standard or deep learning methods. The primary outcomes included pathogenic variant detection performance in 118 cancer-predisposition genes estimated as sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The secondary outcomes were pathogenic variant detection performance in 59 genes deemed actionable by the American College of Medical Genetics and Genomics (ACMG) and 5197 clinically relevant mendelian genes. True sensitivity and true specificity could not be calculated due to lack of a criterion reference standard, but were estimated as the proportion of true-positive variants and true-negative variants, respectively, identified by each method in a reference variant set that consisted of all variants judged to be valid from either approach. The prostate cancer cohort included 1072 men (mean [SD] age at diagnosis, 63.7 [7.9] years; 857 [79.9%] with European ancestry) and the melanoma cohort included 1295 patients (mean [SD] age at diagnosis, 59.8 [15.6] years; 488 [37.7%] women; 1060 [81.9%] with European ancestry). The deep learning method identified more patients with pathogenic variants in cancer-predisposition genes than the standard method (prostate cancer: 198 vs 182; melanoma: 93 vs 74); sensitivity (prostate cancer: 94.7% vs 87.1% [difference, 7.6%; 95% CI, 2.2% to 13.1%]; melanoma: 74.4% vs 59.2% [difference, 15.2%; 95% CI, 3.7% to 26.7%]), specificity (prostate cancer: 64.0% vs 36.0% [difference, 28.0%; 95% CI, 1.4% to 54.6%]; melanoma: 63.4% vs 36.6% [difference, 26.8%; 95% CI, 17.6% to 35.9%]), PPV (prostate cancer: 95.7% vs 91.9% [difference, 3.8%; 95% CI, -1.0% to 8.4%]; melanoma: 54.4% vs 35.4% [difference, 19.0%; 95% CI, 9.1% to 28.9%]), and NPV (prostate cancer: 59.3% vs 25.0% [difference, 34.3%; 95% CI, 10.9% to 57.6%]; melanoma: 80.8% vs 60.5% [difference, 20.3%; 95% CI, 10.0% to 30.7%]). For the ACMG genes, the sensitivity of the 2 methods was not significantly different in the prostate cancer cohort (94.9% vs 90.6% [difference, 4.3%; 95% CI, -2.3% to 10.9%]), but the deep learning method had a higher sensitivity in the melanoma cohort (71.6% vs 53.7% [difference, 17.9%; 95% CI, 1.82% to 34.0%]). The deep learning method had higher sensitivity in the mendelian genes (prostate cancer: 99.7% vs 95.1% [difference, 4.6%; 95% CI, 3.0% to 6.3%]; melanoma: 91.7% vs 86.2% [difference, 5.5%; 95% CI, 2.2% to 8.8%]). Among a convenience sample of 2 independent cohorts of patients with prostate cancer and melanoma, germline genetic testing using deep learning, compared with the current standard genetic testing method, was associated with higher sensitivity and specificity for detection of pathogenic variants. Further research is needed to understand the relevance of these findings with regard to clinical outcomes.

Deep learning has shown promising results in various text-based classification tasks. However, deep learning performance is affected by the number of data, i.e., when the number of data is small, deep learning algorithms do not perform well, and vice versa. Classical machine learning algorithms commonly work well for a few data, and their performance reaches an optimal value and does not increase with the increase in sample data. Therefore, this study aimed to compare the performance of classical machine learning and deep learning methods to detect temperament based on Indonesian Twitter. In this study, the proposed Indonesian Linguistic Inquiry and Word Count were employed to analyze the context of Twitter. The classical machine learning methods implemented were support vector machine and K-nearest neighbor, whereas the deep learning method employed was a convolutional neural network (CNN) with three different architectures. Both learning methods were implemented using multiclass classification and one versus all (OVA) multiclass classification. The highest average f-measure was 58.73%, obtained by CNN OVA with a pool size of 3, a dropout value of 0.7, and a learning rate value of 0.0007.

Speaker recognition is the process of identifying an individual from their voices, and it has been widely applied in many real-world applications. Recently, deep learning has instigated a revolutionary high success rate in speaker recognition. The major advantage of deep learning over conventional methods for speaker recognition is attributed to its representation ability, and the ability to produce highly abstract embedding features from utterances. Recent researches had revealed that deep learning method in learning speaker features from raw data, is strongly depending on a speaker's language. However, only minimal researches had done on deep learning over Vietnamese speaker recognition to present. Nevertheless, this paper has proposed a deep transfer learning method which integrates both transfer learning and deep learning to build models for Vietnamese speaker recognition. Our experimental results indicated that the proposed method is able to build accurate models for Vietnamese speaker recognition.

Electronic mail, or email, is a method for com-municating using the internet which is inexpensive, effective, and fast. Spam is a type of email where unwanted messages, usually unwanted commercial messages, are distributed in large quantities by a spammer. The objective of such behavior is to harm email users; these messages need to be detected and prevented from being sent to users in the first place. In order to filter these emails, the developers have used machine learning methods. This paper discusses different methods which are used deep learning methods such as a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models with(out) a GloVe model in order to classify spam and non-spam messages. These models are only based on email data, and the extraction set of features is automatic. In addition, our work provides a comparison between traditional machine learning and deep learning algorithms on spam datasets to find out the best way to intrusion detection. The results indicate that deep learning offers improved performance of precision, recall, and accuracy. As far as we are aware, deep learning methods show great promise in being able to filter email spam, therefore we have performed a comparison of various deep learning methods with traditional machine learning methods. Using a benchmark dataset consisting of 5,243 spam and 16,872 not-spam and SMS messages, the highest achieved accuracy score is 96.52% using CNN with the GloVe model.

In the setting of primary hyperparathyroidism (PHPT), [18F]fluorocholine PET/CT (FCH-PET) has excellent diagnostic performance, with experienced practitioners achieving 97.7% accuracy in localising hyperfunctioning parathyroid tissue (HPTT). Due to the relative triviality of the task for human readers, we explored the performance of deep learning (DL) methods for HPTT detection and localisation on FCH-PET images in the setting of PHPT. We used a dataset of 93 subjects with PHPT imaged using FCH-PET, of which 74 subjects had visible HPTT while 19 controls had no visible HPTT on FCH-PET. A conventional Resnet10 as well as a novel mPETResnet10 DL model were trained and tested to detect (present, not present) and localise (upper left, lower left, upper right or lower right) HPTT. Our mPETResnet10 architecture also contained a region-of-interest masking algorithm that we evaluated qualitatively in order to try to explain the model's decision process. The models detected the presence of HPTT with an accuracy of 83% and determined the quadrant of HPTT with an accuracy of 74%. The DL methods performed statistically worse (p < 0.001) in both tasks compared to human readers, who localise HPTT with the accuracy of 97.7%. The produced region-of-interest mask, while not showing a consistent added value in the qualitative evaluation of model's decision process, had correctly identified the foreground PET signal. Our experiment is the first reported use of DL analysis of FCH-PET in PHPT. We have shown that it is possible to utilize DL methods with FCH-PET to detect and localize HPTT. Given our small dataset of 93 subjects, results are nevertheless promising for further research.

The clinical utility of regional strain measurements in echocardiography is challenged by suboptimal reproducibility. In this study, we aimed to evaluate the test-retest reproducibility of regional longitudinal strain (RLS) per coronary artery perfusion territory (RLSTerritory) and basal-to-apical level of the left ventricle (RLSLevel), measured by a novel fully automated deep learning (DL) method based on point tracking. We measured strain in a dual-centre test-retest data set that included 40 controls and 40 patients with suspected non-ST elevation acute coronary syndrome. Two consecutive echocardiograms per subject were recorded by different operators. The reproducibility of RLSTerritory and RLSLevel measured by the DL method and by three experienced observers using semi-automatic software (2D Strain, EchoPAC, GE HealthCare) was evaluated as minimal detectable change (MDC). The DL method had MDC for RLSTerritory and RLSLevel ranging from 3.6 to 4.3%, corresponding to a 33-35% improved reproducibility compared with the inter- and intraobserver scenarios (MDC 5.5-6.4% and 4.9-5.4%). Furthermore, the DL method had a lower variance of test-retest differences for both RLSTerritory and RLSLevel compared with inter- and intraobserver scenarios (all P < 0.001). Bland-Altman analyses demonstrated superior reproducibility by the DL method for the whole range of strain values compared with the best observer scenarios. The feasibility of the DL method was 93% and measurement time was only 1 s per echocardiogram. The novel DL method provided fully automated measurements of RLS, with improved test-retest reproducibility compared with semi-automatic measurements by experienced observers. RLS measured by the DL method has the potential to advance patient care through a more detailed, more efficient, and less user-dependent clinical assessment of myocardial function.

Excessive withdrawal of groundwater for agricultural irrigation can cause seawater intrusion into coastal aquifers. Such a case will in turn results in deterioration of irrigation water quality. Determination of irrigation water quality with traditional methods is a time-consuming and costly process. However, machine learning algorithms can be useful tools for modeling and estimating groundwater quality used for irrigation water purposes. In this study, TDS, PS, SAR, and Cl parameters of groundwater were estimated with models based on EC and pH variables. For this purpose, prediction performances of two different deep learning methods (convolutional neural network (CNN) and deep neural network (DNN)) and two different classical machine learning (Random Forest (RF) and extreme gradient boosting (XGBoost)) methods were compared. In addition, predictive uncertainty of the models was determined by quantile regression (QR) analysis. Performance criteria and results of uncertainty analysis revealed that CNN (in testing phase, NSE = 0.95 for TDS, NSE = 0.96 for PS, NSE = 0.67 for SAR and NSE = 0.93 for CI) and DNN (in testing phase, NSE = 0.91 for TDS, NSE = 0.91 for PS, NSE = 0.57 for SAR and NSE = 0.94 for Cl) models had quite a close performance in estimation of TDS, PS, SAR, and Cl parameters and higher than the other two classical machine learning methods. As a result, the CNN model can be considered the best performing model in estimating all quality parameters due to the highest NSE and lowest RMSE values. In addition, the Taylor diagram showed that the values estimated using the CNN model had the highest correlation with the measured data. It was determined that the model with the lowest uncertainty based on the PICP statistics was DNN, followed by the CNN model. However, the CNN model has predicted outliers more accurately. Present findings proved that deep learning models could offer efficient tools for predicting irrigation water quality parameters.

Gastric cancer is the fifth most common cancer in the world. At the same time, it is also the fourth most deadly cancer. Early detection of cancer exists as a guide for the treatment of gastric cancer. Nowadays, computer technology has advanced rapidly to assist physicians in the diagnosis of pathological pictures of gastric cancer. Ensemble learning is a way to improve the accuracy of algorithms, and finding multiple learning models with complementarity types is the basis of ensemble learning. Therefore, this paper compares the performance of multiple algorithms in anticipation of applying ensemble learning to a practical gastric cancer classification problem. The complementarity of sub-size pathology image classifiers when machine performance is insufficient is explored in this experimental platform. We choose seven classical machine learning classifiers and four deep learning classifiers for classification experiments on the GasHisSDB database. Among them, classical machine learning algorithms extract five different image virtual features to match multiple classifier algorithms. For deep learning, we choose three convolutional neural network classifiers. In addition, we also choose a novel Transformer-based classifier. The experimental platform, in which a large number of classical machine learning and deep learning methods are performed, demonstrates that there are differences in the performance of different classifiers on GasHisSDB. Classical machine learning models exist for classifiers that classify Abnormal categories very well, while classifiers that excel in classifying Normal categories also exist. Deep learning models also exist with multiple models that can be complementarity. Suitable classifiers are selected for ensemble learning, when machine performance is insufficient. This experimental platform demonstrates that multiple classifiers are indeed complementarity and can improve the efficiency of ensemble learning. This can better assist doctors in diagnosis, improve the detection of gastric cancer, and increase the cure rate.