AI applications to medical images: From machine learning to deep learning

PurposeArtificial intelligence (AI) models are playing an increasing role in biomedical research and healthcare services. This review focuses on challenges points to be clarified about how to develop AI applications as clinical decision support systems in the real-world context.MethodsA narrative review has been performed including a critical assessment of articles published between 1989 and 2021 that guided challenging sections.ResultsWe first illustrate the architectural characteristics of machine learning (ML)/radiomics and deep learning (DL) approaches. For ML/radiomics, the phases of feature selection and of training, validation, and testing are described. DL models are presented as multi-layered artificial/convolutional neural networks, allowing us to directly process images. The data curation section includes technical steps such as image labelling, image annotation (with segmentation as a crucial step in radiomics), data harmonization (enabling compensation for differences in imaging protocols that typically generate noise in non-AI imaging studies) and federated learning. Thereafter, we dedicate specific sections to: sample size calculation, considering multiple testing in AI approaches; procedures for data augmentation to work with limited and unbalanced datasets; and the interpretability of AI models (the so-called black box issue). Pros and cons for choosing ML versus DL to implement AI applications to medical imaging are finally presented in a synoptic way.ConclusionsBiomedicine and healthcare systems are one of the most important fields for AI applications and medical imaging is probably the most suitable and promising domain. Clarification of specific challenging points facilitates the development of such systems and their translation to clinical practice.     

Artificial intelligence for carbon emissions using system of systems theory

The impact of artificial intelligence (AI) on the environment is the subject of discourse, with arguments for both positive and negative effects. There is a fine line between AI for good and AI for environmental degradation. Today, companies want to seize the benefits of AI, which distinctively involves reducing the company's carbon footprint. However, AI's carbon emissions differ as per the techniques involved in training it. As the saying goes, a coin always has two sides. Therefore, it cannot be denied that AI can be an effective tool for combating climate change, but its role in contributing to carbon emissions cannot be ignored. Multiple studies indicate that AI could be the game-changer in staving off anthropogenic climatic changes due to the deterioration of the environment and global warming. This double-edged relationship and interdependency of AI and carbon emissions are represented through a system of systems (SoS) approach. SoS states that a plan is created through multiple smaller systems, creating complexity in the design and vice versa. A complex system can be assumed as the world in general, where two individual independent systems AI and carbon emissions, when in interaction, create a complex complementary and contradictory relation, adding to the convolution of the system. This connection is demonstrated by conducting a network analysis and calculating the carbon emissions of six machine learning (ML) algorithms and deep learning (DL) models with different datasets but the same hyperparameters on a carbon emission calculator created through AI algorithms. The primary idea of this study is to encourage the AI society to create efficient AI models that may be used without compromising environmental issues. The focus should be on practicing sustainable AI, that is, sustainability from data collection to model deployment, throughout the lifecycle of AI.     

Biocontrol of broomrape (Orobanche crenata Forsk. and Orobanche foetida Poir. ) by Pseudomonas fluorescens isolate Bf7-9 from the faba bean rhizosphere

Bacteria of the faba bean (Vicia faba L.)/Orobanche spp. root environment were evaluated for their potential use as biocontrol agents for the parasitic weed. Bacteria were isolated mainly from the rhizosphere of faba bean as well as from diseased Orobanche underground structures and an Orobanche-suppressive soil from three districts of northern Tunisia. Out of 351 bacterial isolates, 337 were tested for pathogenicity in an inverted pyramidal-shape screening programme including a Lactuca sativa L. seedlings bioassay, root-chamber and pot experiments. In pre-selection screening on L. sativa seedlings, 37 isolates (11%) showed a strong growth inhibitory effect, of which 70 and 84% also had a significant suppressive activity on the pre-emergence structures of O. foetida and O. crenata, respectively, in root-chamber experiments. Among five bacterial isolates selected for pot trials, strain Bf7-9 of Pseudomonas fluorescens showed high biocontrol activity against both species of Orobanche and positively influenced faba bean growth. The bacterium reduced shoot emergence of O. crenata and O. foetida by 64 and 76% and their dry weight by 39 and 63%, respectively, compared with non-inoculated controls. Pseudomonas marginalis strain Nc1-2 exhibited also a tendency to reduce incidence of O. crenata and to improve faba bean performance. Results of the present study suggest that application of naturally occurring rhizosphere bacteria offers an additional approach for biocontrol of Orobanche spp. that can supplement current methods of control in an integrated weed management strategy.     

Artificial intelligence: Deep learning in oncological radiomics and challenges of interpretability and data harmonization

Over the last decade there has been an extensive evolution in the Artificial Intelligence (AI) field. Modern radiation oncology is based on the exploitation of advanced computational methods aiming to personalization and high diagnostic and therapeutic precision. The quantity of the available imaging data and the increased developments of Machine Learning (ML), particularly Deep Learning (DL), triggered the research on uncovering “hidden” biomarkers and quantitative features from anatomical and functional medical images. Deep Neural Networks (DNN) have achieved outstanding performance and broad implementation in image processing tasks. Lately, DNNs have been considered for radiomics and their potentials for explainable AI (XAI) may help classification and prediction in clinical practice. However, most of them are using limited datasets and lack generalized applicability. In this study we review the basics of radiomics feature extraction, DNNs in image analysis, and major interpretability methods that help enable explainable AI. Furthermore, we discuss the crucial requirement of multicenter recruitment of large datasets, increasing the biomarkers variability, so as to establish the potential clinical value of radiomics and the development of robust explainable AI models.     

Integration of A Deep Learning Classifier with A Random Forest Approach for Predicting Malonylation Sites

As a newly-identified protein post-translational modification, malonylation is involved in a variety of biological functions. Recognizing malonylation sites in substrates represents an initial but crucial step in elucidating the molecular mechanisms underlying protein malonylation. In this study, we constructed a deep learning (DL) network classifier based on long short-term memory (LSTM) with word embedding (LSTM_WE) for the prediction of mammalian malonylation sites. LSTM_WE performs better than traditional classifiers developed with common pre-defined feature encodings or a DL classifier based on LSTM with a one-hot vector. The performance of LSTM_WE is sensitive to the size of the training set, but this limitation can be overcome by integration with a traditional machine learning (ML) classifier. Accordingly, an integrated approach called LEMP was developed, which includes LSTM_WE and the random forest classifier with a novel encoding of enhanced amino acid content. LEMP performs not only better than the individual classifiers but also superior to the currently-available malonylation predictors. Additionally, it demonstrates a promising performance with a low false positive rate, which is highly useful in the prediction application. Overall, LEMP is a useful tool for easily identifying malonylation sites with high confidence. LEMP is available at http://www.bioinfogo.org/lemp.     

Investigating genes associated with heart failure,atrial fibrillation,and other cardiovascular diseases,and predicting disease using machine learning techniques for translational research and precision medicine

Cardiovascular disease (CVD) is the leading cause of mortality and loss of disability adjusted life years (DALYs) globally. CVDs like Heart Failure (HF) and Atrial Fibrillation (AF) are associated with physical effects on the heart muscles. As a result of the complex nature, progression, inherent genetic makeup, and heterogeneity of CVDs, personalized treatments are believed to be critical. Rightful application of artificial intelligence (AI) and machine learning (ML) approaches can lead to new insights into CVDs for providing better personalized treatments with predictive analysis and deep phenotyping. In this study we focused on implementing AI/ML techniques on RNA-seq driven gene-expression data to investigate genes associated with HF, AF, and other CVDs, and predict disease with high accuracy. The study involved generating RNA-seq data derived from the serum of consented CVD patients. Next, we processed the sequenced data using our RNA-seq pipeline and applied GVViZ for gene-disease data annotation and expression analysis. To achieve our research objectives, we developed a new Findable, Accessible, Intelligent, and Reproducible (FAIR) approach that includes a five-level biostatistical evaluation, primarily based on the Random Forest (RF) algorithm. During our AI/ML analysis, we have fitted, trained, and implemented our model to classify and distinguish high-risk CVD patients based on their age, gender, and race. With the successful execution of our model, we predicted the association of highly significant HF, AF, and other CVDs genes with demographic variables.     

Gas Exchange and Emission of Chlorophyll Fluorescence during the Monocycle of Rust, Angular Leaf Spot and Anthracnose on Bean Leaves as a Function of their Trophic Characteristics

Measurements related to gas exchange and chlorophyll fluorescence emission were taken from healthy and diseased bean leaves with rust, angular leaf spot, and anthracnose during lesion development for each disease. The experiments were performed at different temperatures of plant incubation, and using two bean cultivars. The main effect of temperature of plant incubation was in disease development. There was no significant difference between cultivars in relation to disease development and in magnitude of physiological alterations when disease severity was the same for each cultivar. These diseases reduced the net photosynthetic rate and increased the dark respiration of infected leaves after the appearance of visible symptoms and the differences between healthy and diseased leaves increased with disease development. The transpiration rate and stomatal conductance were stable during the monocycle of rust, however, these two variables decreased in leaves with angular leaf spot and anthracnose beginning with symptom appearance and continuing until lesion development was complete. Carboxylation resistance was probably the main factor related to reduction of photosynthetic rate of the apparently healthy area of leaves with rust and angular leaf spot. Reduction of the intercellular concentration of CO₂, due to higher stomatal resistance, was probably the main factor for leaves with anthracnose. Chlorophyll fluorescence assessments suggested that there was no change in electron transport capacity and generation of ATP and NADPH in apparently healthy areas of diseased leaves, but decreases in chlorophyll fluorescence emission occurred on visibly lesioned areas for all diseases. Minimal fluorescence was remarkably reduced in leaves with angular leaf spot. Maximal fluorescence and optimal quantum yield of photosystem II of leaves were reduced for all three diseases. Bean rust, caused by a biotrophic pathogen, induced less damage to the regulation mechanisms of the physiological processes of the remaining green area of diseased leaves than did bean angular leaf spot or anthracnose, caused by hemibiotrophic pathogens. The magnitude of photosynthesis reduction can be related to the host–pathogen trophic relationships.     

Feature selection based on empirical-risk function to detect lesions in vascular computed tomography

ObjectiveThe overall goal of the study is to detect coronary artery lesions regardless their nature, calcified or hypo-dense. To avoid explicit modelling of heterogeneous lesions, we adopted an approach based on machine learning and using unsupervised or semi-supervised classifiers. The success of the classifiers based on machine learning strongly depends on the appropriate choice of features differentiating between lesions and regular appearance. The specific goal of this article is to propose a novel strategy devised to select the best feature set for the classifiers used, out of a given set of candidate features.Materials and methodsThe features are calculated in image planes orthogonal to the artery centerline, and the classifier assigns to each of these cross-sections a label “healthy” or “diseased”. The contribution of this article is a feature-selection strategy based on the empirical risk function that is used as a criterion in the initial feature ranking and in the selection process itself. We have assessed this strategy in association with two classifiers based on the density-level detection approach that seeks outliers from the distribution corresponding to the regular appearance. The method was evaluated using a total of 13,687 cross-sections extracted from 53 coronary arteries in 15 patients.ResultsUsing the feature subset selected by the risk-based strategy, balanced error rates achieved by the unsupervised and semi-supervised classifiers respectively were equal to 13.5% and 15.4%. These results were substantially better than the rates achieved using feature subsets selected by supervised strategies. The unsupervised and semi-supervised methods also outperformed supervised classifiers using feature subsets selected by the corresponding supervised strategies.DiscussionSupervised methods require large data sets annotated by experts, both to select the features and to train the classifiers, and collecting these annotations is time-consuming. With these methods, lesions whose appearance differs from the training data may remain undetected. Lesion-detection problem is highly imbalanced, since healthy cross-sections usually are much more numerous than the diseased ones. Training the classifiers based on the density-level detection approach needs a small number of annotations or no annotations at all. The same annotations are sufficient to compute the empirical risk and to perform the selection. Therefore, our strategy associated with an unsupervised or semi-supervised classifier requires a considerably smaller number of annotations as compared to conventional supervised selection strategies. The approach proposed is also better suited for highly imbalanced problems and can detect lesions differing from the training set.ConclusionThe risk-based selection strategy, associated with classifiers using the density-level detection approach, outperformed other strategies and classifiers when used to detect coronary artery lesions. It is well suited for highly imbalanced problems, where the lesions are represented as low-density regions of the feature space, and it can be used in other anomaly detection problems interpretable as a binary classification problem where the empirical risk can be calculated.     

Machine learning for epigenetics and future medical applications

Understanding epigenetic processes holds immense promise for medical applications. Advances in Machine Learning (ML) are critical to realize this promise. Previous studies used epigenetic data sets associated with the germline transmission of epigenetic transgenerational inheritance of disease and novel ML approaches to predict genome-wide locations of critical epimutations. A combination of Active Learning (ACL) and Imbalanced Class Learning (ICL) was used to address past problems with ML to develop a more efficient feature selection process and address the imbalance problem in all genomic data sets. The power of this novel ML approach and our ability to predict epigenetic phenomena and associated disease is suggested. The current approach requires extensive computation of features over the genome. A promising new approach is to introduce Deep Learning (DL) for the generation and simultaneous computation of novel genomic features tuned to the classification task. This approach can be used with any genomic or biological data set applied to medicine. The application of molecular epigenetic data in advanced machine learning analysis to medicine is the focus of this review.     

A novel multi-head CNN design to identify plant diseases using the fusion of RGB images

Plant diseases and insect pests cause a significant threat to agricultural production. Early detection and diagnosis of these diseases are critical and can reduce economic losses. The recent development of deep learning (DL) benefits various fields, such as image processing, remote sensing, medical diagnosis, and agriculture. This work proposed a novel approach based on DL for plant disease detection by fusing RGB and segmented images. A multi-headed DenseNet-based architecture was developed, considering two images as input. We evaluated the model on a public dataset, PlantVillage, consisting of 54183 images with 38 classes. The fivefold cross-validation technique achieved an average accuracy, recall, precision, and f1-score of 98.17%, 98.17%, 98.16%, and 98.12%, respectively. The proposed approach can distinguish various plant diseases with different characteristics by image fusion. The high success rate with low standard deviation proves the robustness of the model, and the model can be integrated into plant disease detection and early warning system.     

Clinical applications of deep learning in breast MRI

Deep learning (DL) is one of the most powerful data-driven machine-learning techniques in artificial intelligence (AI). It can automatically learn from raw data without manual feature selection. DL models have led to remarkable advances in data extraction and analysis for medical imaging. Magnetic resonance imaging (MRI) has proven useful in delineating the characteristics and extent of breast lesions and tumors. This review summarizes the current state-of-the-art applications of DL models in breast MRI. Many recent DL models were examined in this field, along with several advanced learning approaches and methods for data normalization and breast and lesion segmentation. For clinical applications, DL-based breast MRI models were proven useful in five aspects: diagnosis of breast cancer, classification of molecular types, classification of histopathological types, prediction of neoadjuvant chemotherapy response, and prediction of lymph node metastasis. For subsequent studies, further improvement in data acquisition and preprocessing is necessary, additional DL techniques in breast MRI should be investigated, and wider clinical applications need to be explored.     

Forecasting vegetation indices from spatio-temporal remotely sensed data using deep learning-based approaches: A systematic literature review

Over the last few years, Deep learning (DL) approaches have been shown to outperform state-of-the-art machine learning (ML) techniques in many applications such as vegetation forecasting, sales forecast, weather conditions, crop yield prediction, landslides detection and even COVID-19 spread predictions. Several DL algorithms have been employed to facilitate vegetation forecasting research using Remotely Sensed (RS) data. Vegetation is an extremely important component of our global ecosystem and a necessary indicator of land cover dynamics and productivity. Vegetation phenology is influenced by lifecycle patterns, seasonality and weather conditions, leading to changes in their spectral reflectance. Various relevant information, such as vegetation indices (VIs), can be extracted from RS data for vegetation forecasting. Therefore, the Normalized Difference Vegetation Index (NDVI) is known as one of the most widely recognized indices for vegetation related studies. This paper reviews the related works on DL-based spatio-temporal vegetation forecasting using RS data over the period between 2015 and 2021. In this review, we present several DL-based studies and discuss DL algorithms and various sources of data that have been used in these studies. The purpose of this work is to highlight the open challenges such as spatio-temporal prediction issues, spatial and temporal non-stationarity, fusion data, hybrid approaches, deep transfer learning and large parameter requirements. We also attempt to figure out the future directions and limits of DL for vegetation forecasting.     

Sweet potato leaf detection in a natural scene based on faster R-CNN with a visual attention mechanism and DIoU-NMS

Accurate detection of plant leaves is a meaningful and challenging task for developing smart agricultural systems. To improve the performance of detecting plant leaves in natural scenes containing severe occlusion, overlapping, or shape variation, we developed an in situ sweet potato leaf detection method based on a modified Faster R-CNN framework and visual attention mechanism. First, a convolutional block attention module was added to the backbone network to enhance and extract critical features of leaf images by fusing cross-channel information and spatial information. Subsequently, the DIoU-NMS algorithm was adopted to modify the regional proposal network by replacing the original NMS. DIoU-NMS was utilized to reduce missed and incorrect detection in scenes of densely distributed leaves by considering the targets' overlap ratio, distance, and scale. The proposed leaf detection method was tested and evaluated on sweet potato plant images collected in agricultural fields. In the datasets, sweet potato leaves were presented in various sizes and poses, and a large proportion of leaves were occluded or overlapped with each other. The experimental results showed that the proposed leaf detection method outperforms state-of-the-art object detection methods. The mean average precision of the proposed method reached 95.7%, which was 2.9% higher than that of the original Faster R-CNN and 7.0% higher than that of YOLOv5. The proposed method achieved promising performance in detecting dense leaves or occluded leaves and could provide key techniques for applications in smart agriculture and ecological monitoring, such as growth monitoring or plant phenotyping.     

The impact of crop processing on the reconstruction of crop sowing time and cultivation intensity from archaeobotanical weed evidence

Reconstruction of crop sowing time and cultivation intensity, based on arable weed ecology, can resolve archaeological questions surrounding land use and cycles of routine activity, but crop processing may introduce systematic ecological biases in the arable weeds represented in products and by-products. Based on previous ethnoarchaeological work, there is a predicted bias against indicators of spring sowing and intensive cultivation in fine sieve products (and a corresponding over-representation of such species in by-products). Recent work on modern weed floras using functional weed ecology has identified distinctive functional attributes associated with different sowing regimes and cultivation intensity levels. Evaluation of the predicted biases using functional attribute data for modern weed survey studies of different sowing regimes (in Germany) and cultivation intensity levels (in Greece) suggests that there is a likely bias against spring sowing indicators in fine sieve products but not (apparently) against intensive cultivation indicators. An archaeological case study is presented in order to illustrate how bias relating to crop sowing time may be identified and interpreted.     

Machine Learning in Healthcare

Recent advancements in Artificial Intelligence (AI) and Machine Learning (ML) technology have brought on substantial strides in predicting and identifying health emergencies, disease populations, and disease state and immune response, amongst a few. Although, skepticism remains regarding the practical application and interpretation of results from ML-based approaches in healthcare settings, the inclusion of these approaches is increasing at a rapid pace. Here we provide a brief overview of machine learning-based approaches and learning algorithms including supervised, unsupervised, and reinforcement learning along with examples. Second, we discuss the application of ML in several healthcare fields, including radiology, genetics, electronic health records, and neuroimaging. We also briefly discuss the risks and challenges of ML application to healthcare such as system privacy and ethical concerns and provide suggestions for future applications.     

Towards the Disease Biomarker in an Individual Patient Using Statistical Health Monitoring

In metabolomics, identification of complex diseases is often based on application of (multivariate) statistical techniques to the data. Commonly, each disease requires its own specific diagnostic model, separating healthy and diseased individuals, which is not very practical in a diagnostic setting. Additionally, for orphan diseases such models cannot be constructed due to a lack of available data. An alternative approach adapted from industrial process control is proposed in this study: statistical health monitoring (SHM). In SHM the metabolic profile of an individual is compared to that of healthy people in a multivariate manner. Abnormal metabolite concentrations, or abnormal patterns of concentrations, are indicated by the method. Subsequently, this biomarker can be used for diagnosis. A tremendous advantage here is that only data of healthy people is required to construct the model. The method is applicable in current–population based –clinical practice as well as in personalized health applications. In this study, SHM was successfully applied for diagnosis of several orphan diseases as well as detection of metabotypic abnormalities related to diet and drug intake.     

中国传统农业中的生态观及其在技术上的应用

中国传统农业中蕴含着朴素的生态学思想,它以提倡“天人合一”的系统生态观为指导,以精耕细作为特征,以“地力常新壮”为理论,通过物质循环利用,保护自然资源和生物多样性,以及因地、因时、因物制宜地发展农业生产,有力地促进了农业生产力的维持与提高,保证了中华民族的生存与发展。本文对中国传统农业中的生态观及其在技术上的应用范例进行了剖析,并据此提出了中国当代农业在继承传统的基础上进一步发展的途径。     

Improving prediction of chickpea wilt severity using machine learning coupled with model combination techniques under field conditions

Accurate estimation of disease severity in the field is a key to minimize the yield losses in agriculture. Existing disease severity assessment methods have poor accuracy under field conditions. To overcome this limitation, this study used thermal and visible imaging with machine learning (ML) and model combination (MC) techniques to estimate plant disease severity under field conditions. Field experiments were conducted during 2017–18, 2018–19 and 2021–22 to obtain RGB and thermal images of chickpea cultivars with different levels of wilt resistance grown in wilt sick plots. ML models were constructed using four different datasets created using the wilt severity and image derived indices. ML models were also combined using MC techniques to assess the best predictor of the disease severity. Results indicated that the Cubist was the best ML model, while the KNN model was the poorest predictor of chickpea wilt severity under field conditions. MC techniques improved the prediction accuracy of wilt severity over individual ML models. Combining ML models using the least absolute deviation technique gave the best predictions of wilt severity. The results obtained in the present study showed the MC techniques coupled with ML models improved the prediction accuracies of plant disease severity under field conditions.     

Selection of relevant features from amino acids enables development of robust classifiers

Machine learning (ML) has been extensively applied to develop models and to understand high-throughput data of biological processes. However, new ML models, trained with novel experimental results, are required to build regularly for more precise predictions. ML methods can build models from numeric data, whereas biological data are generally textual (DNA, protein sequences) or images and needs feature calculation algorithms to generate quantitative features. Programming skills along with domain knowledge are required to develop these algorithms. Therefore, the process of knowledge discovery through ML is decelerated due to lack of generic tools to construct features and to build models directly from the data. Hence, we developed a schema that calculates about 5,000 features, selects relevant features and develops protein classifiers from the training data. To demonstrate the general applicability and robustness of our method, fungal adhesins and nuclear receptor proteins were used for building classifiers which outperformed existing classifiers when tested on independent data. Next, we built a classifier for mitochondrial proteins of Plasmodium falciparum which causes human malaria because the latest corresponding classifiers are not publically accessible. Our classifier attained 98.18 % accuracy and 0.95 Matthews correlation coefficient by fivefold cross-validation and outperformed existing classifiers on independent test set. We implemented this schema as user-friendly and open source application Pro-Gyan (http://code.google.com/p/pro-gyan/), to build and share executable classifiers without programming knowledge.