Classification Of Types Research Articles

ECG classification via integration of adaptive beat segmentation and relative heart rate with deep learning networks

We propose a state-of-the-art deep learning approach for accurate electrocardiogram (ECG) signal analysis, addressing both waveform delineation and beat type classification tasks. For beat type classification, we integrated two novel schemes into the deep learning model, significantly enhancing its performance. The first scheme is an adaptive beat segmentation method that determines the optimal duration for each heartbeat based on RR-intervals, mitigating segmenting errors from conventional fixed-period segmentation. The second scheme incorporates relative heart rate information of the target beat compared to neighboring beats, improving the model’s ability to accurately detect premature atrial contractions (PACs) that are easily confused with normal beats due to similar morphology. Extensive evaluations on the PhysioNet QT Database, MIT-BIH Arrhythmia Database, and real-world wearable device data demonstrated the proposed approach’s superior capabilities over existing methods in both tasks. The proposed approach achieved sensitivities of 99.81% for normal beats, 99.08% for premature ventricular contractions, and 97.83% for PACs in beat type classification. For waveform delineation, we achieved F1-scores of 0.9842 for non-waveform, 0.9798 for P-waves, 0.9749 for QRS complexes, and 0.9848 for T-waves. It significantly outperforms existing methods in PAC detection while maintaining high performance across both tasks. The integration of aforementioned two schemes into the deep learning model improved the accuracy of normal sinus rhythms and arrhythmia detection.

From Stability to Variability: Classification of Healthy Individuals, Prediabetes, and Type 2 Diabetes Using Glycemic Variability Indices from Continuous Glucose Monitoring Data.

Objective: This study aims to investigate the continuum of glucose control from normoglycemia to dysglycemia (HbA1c ≥ 5.7%/39 mmol/mol) using metrics derived from continuous glucose monitoring (CGM). In addition, we aim to develop a machine learning-based classification model to classify dysglycemia based on observed patterns. Methods: Data from five distinct studies, each featuring at least two days of CGM, were pooled. Participants included individuals classified as healthy, with prediabetes, or with type 2 diabetes mellitus (T2DM). Various CGM indices were extracted and compared across groups. The data set was split 70/30 for training and testing two classification models (XGBoost/Logistic Regression) to differentiate between prediabetes or dysglycemia and the healthy group. Results: The analysis included 836 participants (healthy: n = 282; prediabetes: n = 133; T2DM: n = 432). Across all CGM indices, a progressive shift was observed from the healthy group to those with diabetes (P < 0.001). Statistically significant differences (P < 0.01) were noted in mean glucose, time below range, time above 140 mg/dl, mobility, multiscale complexity index, and glycemic risk index when transitioning from health to prediabetes. The XGBoost models achieved the highest receiver operating characteristic area under the curve values on the test data set ranging from 0.91 [confidence interval (CI): 0.87-0.95] (prediabetes identification) to 0.97 [CI: 0.95-0.98] (dysglycemia identification). Conclusion: Our findings demonstrate a gradual deterioration of glucose homeostasis and increased glycemic variability across the spectrum from normo- to dysglycemia, as evidenced by CGM metrics. The performance of CGM-based indices in classifying healthy individuals and those with prediabetes and diabetes is promising.

An Enhancement for Wireless Body Area Network Using Adaptive Algorithms

Wireless Body Area Networks (WBANs) are one of the most critical technologies for maintaining constant monitoring of patient’s health and diagnosing diseases. They consist of small, wearable wireless sensors transmitting signals. Within this vision, WBANs are not without unique difficulties, for instance, high energy consumption, heat from the sensor, and impaired data accuracy. This paper introduces adaptive algorithms combining Convolutional Neural Networks (CNNs) and dynamic threshold mechanisms to enhance the performance and energy efficiency of Wireless Body Area Networks. The study utilizes the MIB-BIH Arrhythmias dataset to improve the detection of arrhythmias. The results show a 10.53% improvement in battery life and a 5.62-fold enhancement in temperature management when sleep mode technology is applied. As a result, the model reached the average accuracy of ECG classification of 98% and a high level of selectivity and sensitivity to a normal type of heartbeat and quite satisfactory results in the classification of arrhythmia type of heartbeat.

Transformer-based deep learning networks for fault detection, classification, and location prediction in transmission lines

ABSTRACT Fault detection, classification, and location prediction are crucial for maintaining the stability and reliability of modern power systems, reducing economic losses, and enhancing system protection sensitivity. This paper presents a novel Hierarchical Deep Learning Approach (HDLA) for accurate and efficient fault diagnosis in transmission lines. HDLA leverages two-stage transformer-based classification and regression models to perform Fault Detection (FD), Fault Type Classification (FTC), and Fault Location Prediction (FLP) directly from synchronized raw three-phase current and voltage samples. By bypassing the need for feature extraction, HDLA significantly reduces computational complexity while achieving superior performance compared to existing deep learning methods. The efficacy of HDLA is validated on a comprehensive dataset encompassing various fault scenarios with diverse types, locations, resistances, inception angles, and noise levels. The results demonstrate significant improvements in accuracy, recall, precision, and F1-score metrics for classification, and Mean Absolute Errors (MAEs) and Root Mean Square Errors (RMSEs) for prediction, showcasing the effectiveness of HDLA for real-time fault diagnosis in power systems.

Classification of different wheat flour types using hyperspectral imaging and machine learning techniques

Different wheat flour types are used to produce various baked products. Due to the whiteness of the four types, hyperspectral imaging can be used due to receiving infrared wavelength. The technique was applied to distinguish confectionery flour and the flours of Samoun, Sangak, and Tafton breads using a line scanning system in the range of 400–950 nm. Effective wavelengths were selected and different image features were extracted from the corresponding image channels. The selected wavelengths were 601.33, 620.34, 696.41, 730.31, 821.26, and 841.11 nm. The extracted features were used in classification step using linear discriminant analysis, support vector machine, and artificial neural network methods in MATLAB software. The classification accuracy of artificial neural network was higher than the other methods. The efficient features gave higher classification accuracy (98.1 %) than all extracted features (96.9 %). The results showed the high ability of hyperspectral imaging combined with artificial neural network to distinguish different wheat flour types.

Classification and segmentation of five photovoltaic types based on instance segmentation for generating more refined photovoltaic data

Photovoltaic types and spatial information are indispensable for power generation estimation, environmental impact assessment and photovoltaic policy formulation. However, previous studies have predominantly focused on extracting the spatial information of photovoltaics, overlooking the significance of photovoltaic types classification. Moreover, existing research on the classification and segmentation of some novel photovoltaic types is limited. With the widespread adoption of new photovoltaic technologies, the limitations of the data obtained from these studies have become increasingly apparent due to the lack of information on photovoltaic types. To tackle the challenge of the diversification and complexity of photovoltaics, we propose a photovoltaic classification and segmentation network (PV-CSN). This network can automatically classify and segment photovoltaics, providing segmentation results that include information about photovoltaic types. In addition, to enhance network performance, PV-CSN incorporates the receptive field enhancement segment module, BC-PAN-FPN structure, and efficient multi-scale attention module. The experimental results demonstrate the PV-CSN's capability to accurately classify and segment five types of photovoltaics: ground fixed-tilt photovoltaics, ground single-axis tracking photovoltaics, roof photovoltaics, floating water photovoltaics, and stationary water photovoltaics. The Mask-mAP and Box-mAP of this network reach 0.915 and 0.916 respectively, remaining competitive compared to the most advanced instance segmentation networks developed in recent years. These outcomes highlight the superiority of PV-CSN in improving the accuracy of photovoltaic classification and segmentation, as well as solving the identification of new photovoltaics, which provides a promising solution for photovoltaic classification and segmentation tasks.

XgCPred: Cell type classification using XGBoost-CNN integration and exploiting gene expression imaging in single-cell RNAseq data

BackgroundThe advent of single-cell RNA sequencing (scRNA-seq) has revolutionized transcriptomic research by enabling the exploration of gene expression at an individual cell level. This advancement sheds light on how cells differentiate and evolve over time. Effectively classification cell types within scRNA-seq datasets are essential for understanding the intricate cell compositions within tissues and elucidating the origins of various diseases. Challenges persist in the field, emphasizing the need for precise categorization across diverse datasets, addressing aggregated cell data, and managing the complexity of high-dimensional data spaces. MethodologyXgCPred is a novel approach combining XGBoost with convolutional neural networks (CNNs) to provide cell type classification with better accuracy in single-cell RNA-seq data. This combo reveals how well CNNs can detect spatial hierarchy in gene expression images and how XGBoost performs with large volumes of data. XgCPred utilizes an imaging representation of gene expression that is based on the hierarchical organization of genes found in the KEGG BRITE database. ResultsRigorous testing of XgCPred across multiple scRNA-seq datasets, each presenting unique challenges such as varying cell counts, gene expression diversity, and cellular heterogeneity, has demonstrated its superiority compared to earlier methods. The algorithm shows remarkable accuracy and precision in cell type annotation, achieving near-perfect classification scores in some cases. These results underscore its capability to effectively manage data variability. ConclusionsXgCPred distinguishes itself through its dependable and accurate cell type classification across a range of scRNA-seq datasets. Its effectiveness stems from sophisticated data handling and its ability to adapt to the complexities inherent in scRNA-seq data. XgCPred delivers reliable cell annotations essential for further biological analysis and research, marking a significant advancement in genomic studies. With scRNA-seq datasets growing in size and complexity, XgCPred offers a scalable and potent solution for cell type identification, potentially enhancing our understanding of cellular biology and aiding in the precise detection of diseases. XgCPred is a useful tool in genomic research and tailored therapy because it solves current constraints on computing efficiency and generalizability.

AI-driven transcriptomic encoders: From explainable models to accurate, sample-independent cancer diagnostics

In the rapidly evolving domain of medical technology, the utilization of sophisticated algorithms for deciphering transcriptional data has emerged as a critical aspect, especially in the oncology sector. These algorithms, drawing upon methodologies from fields such as natural language processing and advanced image analysis, can significantly enhance the accuracy in predicting cancer-related molecular states. Notably, Transformer models, renowned for their proficiency in handling extensive datasets, are now being adapted for breakthroughs in medical diagnostics or in stratifying patients according to prognostic levels. Our study contributes to the field of precision medicine by integrating Transformer-based learning, exemplified by the Geneformer model, with explainable AI techniques. These techniques are employed to find out the input variables (genes resulting from genomic transcription) most correlated with the decisions of neural network systems. This insight, a key goal in genomic research, aims to select the most relevant gene subset for each specific task in which a neural network is employed. This selection approach has proven to be effective in two classification tasks: cell type classification and breast cancer type classification. Such effectiveness has been demonstrated even across various cohorts of patients. When applying Geneformer-like architecture analyses solely to the selected gene subsets, the outcomes either maintain their accuracy or significantly improve. This approach, aims not only to contribute to the identification of vital genetic markers in cancer genomics, but also to exemplify the adaptability of AI models to different datasets, marking a significant step towards the development of accurate and universally applicable diagnostic tools for precision medicine.

Prediction of amplitude of fault generated travelling wave in medium-voltage grids for fault type classification

This paper proposes a new solution for the classification of short-circuits in medium-voltage distribution grids. The solution requires the measurements of the phase voltages at the time of the short-circuit and the voltage amplitudes of the short-circuit wave resulting from the fault. The values of the voltage before the short-circuit are used to calculate the expected values of the short-circuit wave amplitudes for the different types of short circuits. These calculations are based on an analysis of the charge flow between the capacitances of the power line due to the disturbance. The short-time matrix pencil method was used to detect the incoming short-circuit waves. Tests of the algorithm were carried out on the IEEE 34-bus Feeder model, taking into account the influence of voltage sensors on the measured signal values. The classification algorithm uses only non-zero components due to their relatively low attenuation. Despite that, it achieves high performance in the classification of single- and two-phase short-circuits.

Classification of types of assessment of the efficiency of police activities

The article analyzes the types of assessment of police activity through the prism of the conceptuality of the concept of «effectiveness», the general scientific features of which are: 1) the inherent nature of the category «effectiveness» of an evaluative nature; 2) display with its help the level of achievement of the goal of the activity and taking into account the costs and resources spent for this; 3) implementation of the characterization of the level of efficiency using quantitative and qualitative criteria, the system of which is not stable and may change depending on a number of factors of an objective and subjective nature. It was emphasized that the evaluation of the effectiveness of police activity should be based on the ratio of the obtained result and the costs and resources spent for it. We defined the classification of types of evaluation of the effectiveness of police activity as the process of grouping them according to the signs of kinship and difference. In the presented study, the scientific provisions regarding the classification of types of evaluation of the effectiveness of police activity were further developed. The author’s classification of the types of evaluation of the effectiveness of police activity is proposed, which are divided: depending on the entity that performs the evaluation into state (supra-departmental and departmental) and public; depending on the scale of the national and territorial assessment; depending on the object of assessment for general, special, personal; depending on the content of the statistical and sociological assessment. It is emphasized that the identified criteria are not mutually exclusive, as the evaluation of policing usually involves a combination of different types to obtain a complete comprehensive picture of the effectiveness of policing and may take into account the specific needs and requirements of each region or law enforcement agency. It is noted that the classification of the types of evaluation of the effectiveness of police activity is purely theoretical in nature and allows for a more in-depth clarification of its content and legal nature.

Spatial Immunophenotyping from Whole-Slide Multiplexed Tissue Imaging Using Convolutional Neural Networks.

The multiplexed immunofluorescence (mIF) platform enables biomarker discovery through the simultaneous detection of multiple markers on a single tissue slide, offering detailed insights into intratumor heterogeneity and the tumor-immune microenvironment at spatially resolved single cell resolution. However, current mIF image analyses are labor-intensive, requiring specialized pathology expertise which limits their scalability and clinical application. To address this challenge, we developed CellGate, a deep-learning (DL) computational pipeline that provides streamlined, end-to-end whole-slide mIF image analysis including nuclei detection, cell segmentation, cell classification, and combined immuno-phenotyping across stacked images. The model was trained on over 750,000 single cell images from 34 melanomas in a retrospective cohort of patients using whole tissue sections stained for CD3, CD8, CD68, CK-SOX10, PD-1, PD-L1, and FOXP3 with manual gating and extensive pathology review. When tested on new whole mIF slides, the model demonstrated high precision-recall AUC. Further validation on whole-slide mIF images of 9 primary melanomas from an independent cohort confirmed that CellGate can reproduce expert pathology analysis with high accuracy. We show that spatial immuno-phenotyping results using CellGate provide deep insights into the immune cell topography and differences in T cell functional states and interactions with tumor cells in patients with distinct histopathology and clinical characteristics. This pipeline offers a fully automated and parallelizable computing process with substantially improved consistency for cell type classification across images, potentially enabling high throughput whole-slide mIF tissue image analysis for large-scale clinical and research applications.

Average Nucleotide Identity and Digital DNA-DNA Hybridization Analysis Following PromethION Nanopore-Based Whole Genome Sequencing Allows for Accurate Prokaryotic Typing.

Whole-genome sequencing (WGS) is revolutionizing clinical bacteriology. However, bacterial typing remains investigated by reference techniques with inherent limitations. This stresses the need for alternative methods providing robust and accurate sequence type (ST) classification. This study optimized and evaluated a GridION nanopore sequencing protocol, adapted for the PromethION platform. Forty-eight Escherichia coli clinical isolates with diverse STs were sequenced to assess two alternative typing methods and resistance profiling applications. Multi-locus sequence typing (MLST) was used as the reference typing method. Genomic relatedness was assessed using Average Nucleotide Identity (ANI) and digital DNA-DNA Hybridization (DDH), and cut-offs for discriminative strain resolution were evaluated. WGS-based antibiotic resistance prediction was compared to reference Minimum Inhibitory Concentration (MIC) assays. We found ANI and DDH cut-offs of 99.3% and 94.1%, respectively, which correlated well with MLST classifications and demonstrated potentially higher discriminative resolution than MLST. WGS-based antibiotic resistance prediction showed categorical agreements of ≥ 93% with MIC assays for amoxicillin, ceftazidime, amikacin, tobramycin, and trimethoprim-sulfamethoxazole. Performance was suboptimal (68.8-81.3%) for amoxicillin-clavulanic acid, cefepime, aztreonam, and ciprofloxacin. A minimal sequencing coverage of 12× was required to maintain essential genomic features and typing accuracy. Our protocol allows the integration of PromethION technology in clinical laboratories, with ANI and DDH proving to be accurate and robust alternative typing methods, potentially offering superior resolution. WGS-based antibiotic resistance prediction holds promise for specific antibiotic classes.

Enhancing food crop classification in agriculture through dipper throat optimization and deep learning with remote sensing

Remote sensing images (RSIs), a keystone of modern agricultural technology, refer to spectral or visual data captured from drones, satellites, or aircraft without direct physical contact with the Earth's surface. These images provide a wide-ranging view of agricultural landscapes, providing valuable insights into land use, crop health, and environmental conditions. Agricultural food crop classification, a vital application within precision agriculture, includes the detection and classification of different crops cultivated in a certain region. Traditionally reliant on manual techniques, the development of technologies, particularly the incorporation of RSIs, has revolutionized this process. Agricultural food crop classification has become more sophisticated and automated by harnessing the wealth of data received from RS, which facilitates precise management and monitoring of crops on a large scale. Deep learning (DL), a branch of artificial intelligence, plays a more effective role in these synergies. The incorporation of DL into the RSI analysis enables high-precision and efficient detection of various crop types, assisting more informed decision-making in agriculture. This study proposes a new Dipper Throat Optimization Algorithm with Deep Learning based Food Crop Classification (DTOADL-FCC) algorithm using Remote Sensing Imaging for Agricultural Resource Management. The DTOADL-FCC method aims to apply DL algorithms for the classification of different crop types. In the DTOADL-FCC method, fully convolutional network (FCN) based segmentation process is performed. Next, the DTOADL-FCC method exploits the SE-ResNet model for learning intrinsic and complex features. The DTOADL-FCC method makes use of DTOA for the hyperparameter tuning process. Lastly, the classification of crop types takes place using the extreme learning machine (ELM) model. The study utilizes mathematical formulations including activation functions, loss functions, fitness calculations, and iterative update processes. A brief set of simulations showcases that the DTOADL-FCC method achieves remarkable performance over other techniques with much improved results.

Segmentation of ovarian cyst in ultrasound images using AdaResU-net with optimization algorithm and deep learning model

Ovarian cysts pose significant health risks including torsion, infertility, and cancer, necessitating rapid and accurate diagnosis. Ultrasonography is commonly employed for screening, yet its effectiveness is hindered by challenges like weak contrast, speckle noise, and hazy boundaries in images. This study proposes an adaptive deep learning-based segmentation technique using a database of ovarian ultrasound cyst images. A Guided Trilateral Filter (GTF) is applied for noise reduction in pre-processing. Segmentation utilizes an Adaptive Convolutional Neural Network (AdaResU-net) for precise cyst size identification and benign/malignant classification, optimized via the Wild Horse Optimization (WHO) algorithm. Objective functions Dice Loss Coefficient and Weighted Cross-Entropy are optimized to enhance segmentation accuracy. Classification of cyst types is performed using a Pyramidal Dilated Convolutional (PDC) network. The method achieves a segmentation accuracy of 98.87%, surpassing existing techniques, thereby promising improved diagnostic accuracy and patient care outcomes.

Influence of Fractures and Dislocations in Severity of Rocker-Bottom Deformity in Patients with Charcot Foot.

The objective of this study was to identify bone fractures and joint dislocations that have greater association with the severity of arch collapse in patients with Charcot foot involving the midfoot.A retrospective study in 28 (N = 29 feet) patients who had Charcot foot deformity of the midfoot. The study included stage III of Eichenholtz classification, and Schon classification types I to III. Talar-first metatarsal and calcaneal pitch angles and cuboid height were used to evaluate the severity of the midfoot deformity in a weightbearing lateral radiograph. Two investigators evaluated the bone fracture and joint dislocation involved in weightbearing antero-posterior and lateral radiographs.There were 13 (46%) feet that showed pattern 1, 9 (31%) feet with pattern 2, and 7 (25%) feet with pattern 3 according to the Schon classification. One foot had a combination of patterns 1 and 2. Midfoot ulceration occurred in 64% (n = 19) of feet. In the multivariate analysis, plantarflexion of talar-first metatarsal angle was predicted by navicular-medial cuneiform dislocation (p = .007 [-20.620-3.683]), an increase of the negative calcaneal pitch angle by fragmentation of the cuboid (p = .003 [-15.568-3.626]), and increment of the negative cuboid height by navicular-medial cuneiform and medial cuneiform-first metatarsal dislocations (p = .040 [-12.779-0.317], p = .002 [-13.437-3.267], respectively).Bone fractures and dislocations in the sagittal plane seem to contribute to midfoot collapse, but navicular-medial cuneiform dislocation/non-union and cuboid fragmentation predict severe rocker-bottom deformity in cases of Charcot foot.

ХИМИЯЛЫК ЖЕР СЕМИРТКИЧТЕРДИ АЙЫЛ ЧАРБАСЫНДА ПАЙДАЛАНУУНУН КООПСУЗДУГУ

This article talks about chemical fertilizers and provides safety precautions when using them. It also shows the classification of fertilizers, types, safety measures and negative consequences, as well as ways to solve them. Today it is noted that research is being conducted on the impact of chemicalization on agriculture, the environment and human health. Chemical fertilizers are widely used in agriculture to increase crop yields. They provide nutrients to plants, improve soil structure, and help reduce plant diseases and pests. However, improper and excessive use of chemical fertilizers can have negative impacts on the environment and human health. This article discusses the safe use of chemical fertilizers and their effects.

Human rights education as a human right – A logical analysis based on Kanger’s theory of rights

This article provides a theoretical foundation to argue for an effective and meaningful realization of human rights education in school curricula and beyond. Despite emerging research on a content-related basis and large-scale initiatives, the practical realization of human rights education is mostly still a desideratum. By analyzing relevant documents by the United Nations (UN), particularly the Universal Declaration of Human Rights (UNGA, 1948), this paper shows that human rights education is a human right itself. Upon joining the UN, member states pledged to provide human rights education to all citizens. However, the question what it means to ensure the right to human rights education is complex because the concept of ‘right’ is ambiguous. Thus, this manuscript logically analyzes the right to human rights education by means of Kanger’s theory of rights that encompasses a classification of all possible types of right. The analysis shows that the right to human rights education consists of two complex types of right that contain both a claim right as well as service rights. The author concludes that in order to meet the obligations of the UN Charta, UN member states are required (1) to provide mandatory human rights education in compulsory education and (2) to grant access to human rights education to all citizens.

Explainable Feature Engineering for Multi-Modal Tissue State Monitoring Based on Impedance Spectroscopy.

One of the most promising approaches to food quality assessments is the use of impedance spectroscopy combined with machine learning. Thereby, feature selection is decisive for a high classification accuracy. Physically based features have particularly significant advantages because they are able to consider prior knowledge and to concentrate the data into pertinent understandable information, building a solid basis for classification. In this study, we aim to identify physically based measurable features for muscle type and freshness classifications of bovine meat based on impedance spectroscopy measurements. We carry out a combined study where features are ranked based on their F1-score, cumulative feature selection, and t-distributed Stochastic Neighbor Embedding (t-SNE). In terms of features, we analyze the characteristic points (CPs) of the impedance spectrum and the model parameters (MPs) obtained by fitting a physical model to the measurements. The results show that either MPs or CPs alone are sufficient for detecting muscle type. Combining capacitance (C) and extracellular resistance (Rex) or the modulus of the characteristic point Z1 and the phase at the characteristic frequency of the beta dispersion (Phi2) leads to accurate separation. In contrast, the detection of freshness is more challenging. It requires more distinct features. We achieved a 90% freshness separation using the MPs describing intracellular resistance (Rin) and capacitance (C). A 95.5% freshness separation was achieved by considering the phase at the end of the beta dispersion (Phi3) and Rin. Including additional features related to muscle type improves the separability of samples; ultimately, a 99.6% separation can be achieved by selecting the appropriate features.

Metadata-enhanced contrastive learning from retinal optical coherence tomography images

Deep learning has potential to automate screening, monitoring and grading of disease in medical images. Pretraining with contrastive learning enables models to extract robust and generalisable features from natural image datasets, facilitating label-efficient downstream image analysis. However, the direct application of conventional contrastive methods to medical datasets introduces two domain-specific issues. Firstly, several image transformations which have been shown to be crucial for effective contrastive learning do not translate from the natural image to the medical image domain. Secondly, the assumption made by conventional methods, that any two images are dissimilar, is systematically misleading in medical datasets depicting the same anatomy and disease. This is exacerbated in longitudinal image datasets that repeatedly image the same patient cohort to monitor their disease progression over time. In this paper we tackle these issues by extending conventional contrastive frameworks with a novel metadata-enhanced strategy. Our approach employs widely available patient metadata to approximate the true set of inter-image contrastive relationships. To this end we employ records for patient identity, eye position (i.e. left or right) and time series information. In experiments using two large longitudinal datasets containing 170,427 retinal optical coherence tomography (OCT) images of 7912 patients with age-related macular degeneration (AMD), we evaluate the utility of using metadata to incorporate the temporal dynamics of disease progression into pretraining. Our metadata-enhanced approach outperforms both standard contrastive methods and a retinal image foundation model in five out of six image-level downstream tasks related to AMD. We find benefits in both a low-data and high-data regime across tasks ranging from AMD stage and type classification to prediction of visual acuity. Due to its modularity, our method can be quickly and cost-effectively tested to establish the potential benefits of including available metadata in contrastive pretraining.

Turning the attention to time-resolved EPID-images: treatment error classification with transformer multiple instance learning

Objective. The aim of this work was to develop a novel artificial intelligence-assisted in vivo dosimetry method using time-resolved (TR) dose verification data to improve quality of external beam radiotherapy. Approach. Although threshold classification methods are commonly used in error classification, they may lead to missing errors due to the loss of information resulting from the compression of multi-dimensional electronic portal imaging device (EPID) data into one or a few numbers. Recent research has investigated the classification of errors on time-integrated (TI) in vivo EPID images, with convolutional neural networks showing promise. However, it has been observed previously that TI approaches may cancel out the error presence on γ-maps during dynamic treatments. To address this limitation, simulated TR γ-maps for each volumetric modulated arc radiotherapy angle were used to detect treatment errors caused by complex patient geometries and beam arrangements. Typically, such images can be interpreted as a set of segments where only set class labels are provided. Inspired by recent weakly supervised approaches on histopathology images, we implemented a transformer based multiple instance learning approach and utilized transfer learning from TI to TR γ-maps. Main results. The proposed algorithm performed well on classification of error type and error magnitude. The accuracy in the test set was up to 0.94 and 0.81 for 11 (error type) and 22 (error magnitude) classes of treatment errors, respectively. Significance. TR dose distributions can enhance treatment delivery decision-making, however manual data analysis is nearly impossible due to the complexity and quantity of this data. Our proposed model efficiently handles data complexity, substantially improving treatment error classification compared to models that leverage TI data.