Achieved Accuracy Levels Research Articles

DeepEMPR: coffee leaf disease detection with deep learning and enhanced multivariance product representation.

Plant diseases threaten agricultural sustainability by reducing crop yields. Rapid and accurate disease identification is crucial for effective management. Recent advancements in artificial intelligence (AI) have facilitated the development of automated systems for disease detection. This study focuses on enhancing the classification of diseases and estimating their severity in coffee leaf images. To do so, we propose a novel approach as the preprocessing step for the classification in which enhanced multivariance product representation (EMPR) is used to decompose the considered image into components, a new image is constructed using some of those components, and the contrast of the new image is enhanced by applying high-dimensional model representation (HDMR) to highlight the diseased parts of the leaves. Popular convolutional neural network (CNN) architectures, including AlexNet, VGG16, and ResNet50, are evaluated. Results show that VGG16 achieves the highest classification accuracy of approximately 96%, while all models perform well in predicting disease severity levels, with accuracies exceeding 85%. Notably, the ResNet50 model achieves accuracy levels surpassing 90%. This research contributes to the advancement of automated crop health management systems.

Regulatory‐based classification of rums: a chemometric and machine learning analysis

SummaryThe Industria Licorera de Caldas (ILC) stands as a major liquor factory in Colombia, specialising in the production of various rum types including Tradicional, Juan de la Cruz, Carta de Oro, and Reserva Especial. These rums, as congeneric drinks, are known for their rich content of volatile compounds that define their sensory characteristics. To be commercialised, each rum batch must comply with Colombian standard NTC278 which defines rigorous assessment of congener content and various physicochemical parameters. Thus, the ILC has accumulated a vast amount of data over the years. This study conducts a comprehensive analysis of ILC rums, using chemometric techniques and machine‐learning classification models such as PCA, KNN, LDA, and RF. The aim was to distinguish between rum types based on parameters specified for standard compliance, streamlining the process without the need for additional or extensive new methodologies. As a result, through PCA data exploration, it was revealed that acetaldehyde, ethyl acetate, and isobutanol levels are instrumental in differentiating rum variants. Similarly, all classification models achieved accuracy levels exceeding 0.83 and precision surpassing 0.93. These findings pave the way for further research in the development of an ILC‐specific sensor for rapid and reliable liquor authenticity testing.

Deep Learning‐Based Automated Detection and Grading of Papilledema From OCT Images: A Promising Approach for Improved Clinical Diagnosis and Management

ABSTRACTPapilledema is a prevalent neuro‐ophthalmic condition characterized by optic disk swelling. It is known to pose a significant risk of vision loss in its advanced stages. To address the pressing need for accurate detection and grading of papilledema, this study introduces a novel approach utilizing optical coherence tomography (OCT) scans. A cascaded model that combines four transfer learning models—SqueezeNet, AlexNet, GoogleNet, and ResNet‐50—for both the detection and grading phases was proposed. Additionally, a specialized convolutional neural network (CNN) model is meticulously designed to cater specifically to the complexities of papilledema analysis. Unlike the fundus camera‐based models, this study integrates deep learning models for the diagnosis of papilledema from OCT scans. A new dataset of OCT scans was collected to ensure a comprehensive evaluation of the models. It encompasses a wide range of papilledema, pseudopapilledema, and normal cases. This dataset serves as a valuable resource for training and testing of the proposed models. In addition, two validation strategies have been adopted to ensure the model's generalizability and robustness. Furthermore, it enhances the model's accuracy and reliability. The results are highly promising; remarkable accuracy rates have been achieved. Specifically, the SqueezeNet, AlexNet, GoogleNet, ResNet‐50, and customized CNN models achieved accuracy levels of 98.44%, 98.50%, 98.28%, 98.30%, and 96.26%, respectively, for the handout validation strategy. These findings not only demonstrate the efficacy of using deep learning in papilledema detection and grading but also establish the superiority of the proposed models when compared with other relevant studies. By addressing the challenges associated with papilledema, the study significantly contributes to the advancement of neuro‐ophthalmic diagnostics. The accurate and efficient detection of papilledema from OCT scans holds immense potential for guiding timely interventions and preserving patients' visual health.

COMPARATIVE ANALYSIS OF LOW-COST CONSUMER-GRADE AND COMMERCIAL-GRADE DRONES FOR LAND SURVEYING APPLICATIONS

Abstract. This study investigates the feasibility of using low-cost consumer drones, specifically the DJI Mini 2, for land surveys compared to commercial-grade drones like the DJI Phantom 4 Pro V2. The research, conducted within the University of the Philippines - Diliman campus, evaluates the horizontal accuracy of resulting ortho-images and explores the advantages and disadvantages of each platform. Initiated by the recognition of aerial drones' transformative impact on land surveying, particularly acknowledged by the Land Management Bureau in the Philippines, the study addresses concerns about the use of consumer-grade drones for surveying. While some advocate for commercial-grade drones, citing concerns about accuracy, a prior study in India demonstrated the suitability of consumer-grade drones for geomatics applications. The methodology involves creating a flight plan with specified parameters for both drones, establishing Ground Control Points (GCPs), and conducting two flights. The results, processed using UAS Mapping Workflow, indicate comparable errors between the consumer-grade DJI Mini 2 and commercial-grade DJI Phantom 4 Pro V2. Despite slight differences, both drones meet the accuracy standards for land surveys. The study contributes to the discourse on affordable means of secure land tenure, envisioning consumer-grade drones as viable alternatives for cost-effective land surveys. The DJI Mini 2, at a fraction of the cost, demonstrates the capability to achieve accuracy levels comparable to its commercial counterpart. While acknowledging the need for further testing and optimization, the findings suggest the potential of consumer-grade drones in advancing sustainable urban development and poverty reduction through accessible land survey methods.

Integration of lanthanide MOFs/methylcellulose-based fluorescent sensor arrays and deep learning for fish freshness monitoring

Preserving fish meat poses a significant challenge due to its high protein and low fat content. This study introduces a novel approach that utilizes a common type of lanthanide metal-organic frameworks (Ln-MOFs), EuMOFs, in combination with 5-fluorescein isothiocyanate (FITC) and methylcellulose (MC) to develop fluorescent sensor arrays for real-time monitoring the freshness of fish meat. The EuMOF-FITC/MC fluorescence films were characterized with excellent fluorescence response, ideal morphology, good mechanical properties, and improved hydrophobicity. The efficacy of the fluorescence sensor array was evaluated by testing various concentrations of spoilage gases (such as ammonia, dimethylamine, and trimethylamine) within a 20-min timeframe using a smartphone-based camera obscura device. This sensor array enables the real-time monitoring of fish freshness, with the ability to preliminarily identify the freshness status of mackerel meat with the naked eye. Furthermore, the study employed four convolutional neural network (CNN) models to enhance the performance of freshness assessment, all of which achieved accuracy levels exceeding 93 %. Notably, the ResNext-101 model demonstrated a particularly high accuracy of 98.97 %. These results highlight the potential of the EuMOF-based fluorescence sensor array, in conjunction with the CNN model, as a reliable and accurate method for real-time monitoring the freshness of fish meat.

Deep transformer-based personalized dosimetry from SPECT/CT images: a hybrid approach for [177Lu]Lu-DOTATATE radiopharmaceutical therapy

PurposeAccurate dosimetry is critical for ensuring the safety and efficacy of radiopharmaceutical therapies. In current clinical dosimetry practice, MIRD formalisms are widely employed. However, with the rapid advancement of deep learning (DL) algorithms, there has been an increasing interest in leveraging the calculation speed and automation capabilities for different tasks. We aimed to develop a hybrid transformer-based deep learning (DL) model that incorporates a multiple voxel S-value (MSV) approach for voxel-level dosimetry in [177Lu]Lu-DOTATATE therapy. The goal was to enhance the performance of the model to achieve accuracy levels closely aligned with Monte Carlo (MC) simulations, considered as the standard of reference. We extended our analysis to include MIRD formalisms (SSV and MSV), thereby conducting a comprehensive dosimetry study.MethodsWe used a dataset consisting of 22 patients undergoing up to 4 cycles of [177Lu]Lu-DOTATATE therapy. MC simulations were used to generate reference absorbed dose maps. In addition, MIRD formalism approaches, namely, single S-value (SSV) and MSV techniques, were performed. A UNEt TRansformer (UNETR) DL architecture was trained using five-fold cross-validation to generate MC-based dose maps. Co-registered CT images were fed into the network as input, whereas the difference between MC and MSV (MC-MSV) was set as output. DL results are then integrated to MSV to revive the MC dose maps. Finally, the dose maps generated by MSV, SSV, and DL were quantitatively compared to the MC reference at both voxel level and organ level (organs at risk and lesions).ResultsThe DL approach showed slightly better performance (voxel relative absolute error (RAE) = 5.28 ± 1.32) compared to MSV (voxel RAE = 5.54 ± 1.4) and outperformed SSV (voxel RAE = 7.8 ± 3.02). Gamma analysis pass rates were 99.0 ± 1.2%, 98.8 ± 1.3%, and 98.7 ± 1.52% for DL, MSV, and SSV approaches, respectively. The computational time for MC was the highest (~2 days for a single-bed SPECT study) compared to MSV, SSV, and DL, whereas the DL-based approach outperformed the other approaches in terms of time efficiency (3 s for a single-bed SPECT). Organ-wise analysis showed absolute percent errors of 1.44 ± 3.05%, 1.18 ± 2.65%, and 1.15 ± 2.5% for SSV, MSV, and DL approaches, respectively, in lesion-absorbed doses.ConclusionA hybrid transformer-based deep learning model was developed for fast and accurate dose map generation, outperforming the MIRD approaches, specifically in heterogenous regions. The model achieved accuracy close to MC gold standard and has potential for clinical implementation for use on large-scale datasets.

Diagnosis of Coronary Artery Disease using Adult Data from Blood Tests and Electrocardiograms

Objective:
 Many modifiable risk factors affect the onset of coronary artery disease (CAD), a condition that is extremely common throughout the globe. Predictive models created using machine learning (ML) algorithms may help physicians identify CAD earlier and may lead to better results. The goal of this project was to use ML algorithms to predict CAD in patients.
 
 Methods:
 The gathered dataset of UCI heart disease was used in this study to evaluate a variety of machine learning methods to predict CAD. Just the most crucial aspects of the hypothesis testing method were kept. Support vector machines (SVM) were used in a comparative analysis employing a variety of assessment measures.
 
 Results:
 All machine learning methods achieved accuracy levels of at least 80%, with the SVM algorithm obtaining accuracy levels of at least 90%. Predictive ML models had high diagnostic relevance in CAD, as seen by the SVM model's high recall (0.9), which is was the highest of all the models.
 
 Conclusion:
 The findings of the current study demonstrated that, independent of the measures used to evaluate machine learning models, feature selection has a significant impact on performance. Finding the most useful features is thus crucial. SVM was chosen as the top model based on the features we considered.

Predicting ferry services with integrated meteorological data using machine learning

Ferry services that connect a huge number of islands and mainlands are vital transportation methods in several nations. However, a major disadvantage of ferry services is that they are crucially affected by weather conditions. Informing customers about regular ferry service operations is thus very important. With this in mind, the aim of this study was to predict whether ferry services can be provided in a timely manner through machine learning approaches with meteorological (6–48 h prior) and operation data sets. It was found that the random forest classifier achieved accuracy levels of 90.50% (6 h prior) and 88.78% (48 h prior) in predicting ferry services, which were greater than regulation-oriented determination. Both implications and limitations are presented based on the findings of this study.

Empowering Deaf-Hearing Communication: Exploring Synergies between Predictive and Generative AI-Based Strategies towards (Portuguese) Sign Language Interpretation.

Communication between Deaf and hearing individuals remains a persistent challenge requiring attention to foster inclusivity. Despite notable efforts in the development of digital solutions for sign language recognition (SLR), several issues persist, such as cross-platform interoperability and strategies for tokenizing signs to enable continuous conversations and coherent sentence construction. To address such issues, this paper proposes a non-invasive Portuguese Sign Language (Língua Gestual Portuguesa or LGP) interpretation system-as-a-service, leveraging skeletal posture sequence inference powered by long-short term memory (LSTM) architectures. To address the scarcity of examples during machine learning (ML) model training, dataset augmentation strategies are explored. Additionally, a buffer-based interaction technique is introduced to facilitate LGP terms tokenization. This technique provides real-time feedback to users, allowing them to gauge the time remaining to complete a sign, which aids in the construction of grammatically coherent sentences based on inferred terms/words. To support human-like conditioning rules for interpretation, a large language model (LLM) service is integrated. Experiments reveal that LSTM-based neural networks, trained with 50 LGP terms and subjected to data augmentation, achieved accuracy levels ranging from 80% to 95.6%. Users unanimously reported a high level of intuition when using the buffer-based interaction strategy for terms/words tokenization. Furthermore, tests with an LLM-specifically ChatGPT-demonstrated promising semantic correlation rates in generated sentences, comparable to expected sentences.

Segmentation quality assessment by automated detection of erroneous surface regions in medical images

Despite the advancement in deep learning-based semantic segmentation methods, which have achieved accuracy levels of field experts in many computer vision applications, the same general approaches may frequently fail in 3D medical image segmentation due to complex tissue structures, noisy acquisition, disease-related pathologies, as well as the lack of sufficiently large datasets with associated annotations. For expeditious diagnosis and quantitative image analysis in large-scale clinical trials, there is a compelling need to predict segmentation quality without ground truth. In this paper, we propose a deep learning framework to locate erroneous regions on the boundary surfaces of segmented objects for quality control and assessment of segmentation. A Convolutional Neural Network (CNN) is explored to learn the boundary related image features of multi-objects that can be used to identify location-specific inaccurate segmentation. The predicted error locations can facilitate efficient user interaction for interactive image segmentation (IIS). We evaluated the proposed method on two data sets: Osteoarthritis Initiative (OAI) 3D knee MRI and 3D calf muscle MRI. The average sensitivity scores of 0.95 and 0.92, and the average positive predictive values of 0.78 and 0.91 were achieved, respectively, for erroneous surface region detection of knee cartilage segmentation and calf muscle segmentation. Our experiment demonstrated promising performance of the proposed method for segmentation quality assessment by automated detection of erroneous surface regions in medical images.

A multimodal dual-fusion entity extraction model for large and complex devices

In the context of large and complex devices with a multi-source heterogeneous data environment, the extraction of network device configuration related entity information from diverse modalities of the Internet of Things data is a crucial and fundamental step towards establishing a domain knowledge graph for global Zero Touch Provisioning of network devices. In this paper, we present a novel multimodal dual-fusion entity extraction model that serves as a foundation for an intelligent and efficient network device configuration process. Firstly, the multimodal data is encoded, followed by the ViLBERT pre-training model to obtain more feature information of entities for multimodal front-end fusion. Next, the attention weights of each modal feature are learned through a multilayer neural network classifier and probabilistic graphical model, facilitating multimodal back-end fusion and reducing information redundancy. Finally, entity recognition is accomplished by employing a cohesive memory module that extracts the essential parameters for device configuration. The simulation results demonstrate that the proposed model performs exceptionally well on the MSCOCO2017 public dataset and the SFZK-Dev data network device dataset, with F1 values of the comprehensive evaluation index of model quality at 94.65% and 96.94%, respectively, indicating high stability. Additionally, the link prediction metrics HITS@1 achieved accuracy levels of 58.21% and 68.04%.

Research on crop classification methods based on machine learning using wavelet transformations

Due to the growing demand for precise and efficient agricultural monitoring and management systems, interest in crop classification using machine learning has significantly increased in recent years. Conventional machine learning algorithms, however, have limitations when dealing with high-dimensional data. Training time is a crucial factor in developing crop classification models, as it directly impacts the model's efficacy and efficiency. Large volumes of data are often needed to train crop classification models properly, making the training procedure laborious and computationally demanding. In this paper, crop classification model training time is reduced by utilizing wavelet decomposition in combination with traditional machine learning techniques such as SVM, Naive Bayes, RF, and others. The performance of different algorithms before and after utilizing wavelet decomposition was evaluated in order to find the way that is the most efficient while using this methodology. Additionally, the significance of quality loss when applying wavelet coefficients was determined. The results of this paper show that applying wavelet transformation coefficients in combination with classification techniques can achieve accuracy levels that are comparable to those achieved by training on the original images. For example, using the Random Forest model in combination with Daubechies transformation coefficients can achieve an accuracy of 0,89 while significantly reducing training time from 11,15 to 3,49 seconds with Haar transformation providing almost identical results. The paper demonstrates the value of using wavelet transforms for crop classification and highlights the significance of accounting for training time when developing accurate and practical crop classification models that may be useful in developing decision support tools for agricultural applications, where it is crucial to make prompt decisions based on current data.

Compact-Fusion Feature Framework for Ethnicity Classification

In computer vision, ethnicity classification tasks utilize images containing human faces to extract ethnicity labels. Ethnicity is one of the soft biometric feature categories useful in data analysis for commercial, public, and health sectors. Ethnicity classification begins with face detection as a preprocessing process to determine a human’s presence; then, the feature representation is extracted from the isolated facial image to predict the ethnicity class. This study utilized four handcrafted features (multi-local binary pattern (MLBP), histogram of gradient (HOG), color histogram, and speeded-up-robust-features-based (SURF-based)) as the basis for the generation of a compact-fusion feature. The compact-fusion framework involves optimal feature selection, compact feature extraction, and compact-fusion feature representation. The final feature representation was trained and tested with the SVM One Versus All classifier for ethnicity classification. When it was evaluated in two large datasets, UTKFace and Fair Face, the proposed framework achieved accuracy levels of 89.14%, 82.19%, and 73.87%, respectively, for the UTKFace dataset with four or five classes and the Fair Face dataset with four classes. Furthermore, the compact-fusion feature with a small number of features at 4790, constructed based on conventional handcrafted features, achieved competitive results compared with state-of-the-art methods using a deep-learning-based approach.

Unified Framework for Identity and Imagined Action Recognition From EEG Patterns

We present a unified deep learning framework for the recognition of user identity and the recognition of imagined actions, based on electroencephalography (EEG) signals, for application as a brain–computer interface. Our solution exploits a novel shifted subsampling preprocessing step as a form of data augmentation, and a matrix representation to encode the inherent local spatial relationships of multielectrode EEG signals. The resulting image-like data are then fed to a convolutional neural network to process the local spatial dependencies, and eventually analyzed through a bidirectional long-short term memory module to focus on temporal relationships. Our solution is compared against several methods in the state of the art, showing comparable or superior performance on different tasks. Specifically, we achieve accuracy levels above 90% both for action and user classification tasks. In terms of user identification, we reach 0.39% equal error rate in the case of known users and gestures, and 6.16% in the more challenging case of unknown users and gestures. Preliminary experiments are also conducted in order to direct future works toward everyday applications relying on a reduced set of EEG electrodes.

Network Monitoring on Multi-Pipe Switches

Programmable switches have been widely used to design network monitoring solutions that operate in the fast data-plane level, e.g., detecting heavy hitters, super-spreaders, computing flow size distributions and their entropy. Many existing works on networking monitoring assume switches deploy a single memory that is accessible by each processed packet. However, high-speed ASIC switches increasingly deploymultiple independent pipes, each equipped with its own independent memory thatcannot be accessed by other pipes. In this work, we initiate the study of deploying existing heavy-hitter data-plane monitoring solutions on multi-pipe switches where packets of a "flow" may spread over multiple pipes, i.e., stored into distinct memories. We first quantify the accuracy degradation due to splitting a monitoring data structure across multiple pipes (e.g., up to 3000x worse flow-size estimation average error). We then present PipeCache, a system that adaptsexisting data-plane mechanisms to multi-pipe switches by carefully storing all the monitoring information of each traffic class into exactly one specific pipe (as opposed to replicate the information on multiple pipes). PipeCache relies on the idea of briefly storing monitoring information into a per-pipe cache and then piggybacking this information onto existing data packets to the correct pipeentirely at data-plane speed. We implement PipeCache on ASIC switches and we evaluate it using a real-world trace. We show that existing data-plane mechanisms achieves accuracy levels and memory requirements similar to single-pipe deployments when augmented with PipeCache (i.e., up to 16x lower memory requirements).

An Accurate Deep Learning–Based System for Automatic Pill Identification: Model Development and Validation

Medication errors account for a large proportion of all medical errors. In most homes, patients take a variety of medications for a long period. However, medication errors frequently occur because patients often throw away the containers of their medications. We proposed a deep learning-based system for reducing medication errors by accurately identifying prescription pills. Given the pill images, our system located the pills in the respective pill databases in South Korea and the United States. We organized the system into a pill recognition step and pill retrieval step, and we applied deep learning models to train not only images of the pill but also imprinted characters. In the pill recognition step, there are 3 modules that recognize the 3 features of pills and their imprints separately and correct the recognized imprint to fit the actual data. We adopted image classification and text detection models for the feature and imprint recognition modules, respectively. In the imprint correction module, we introduced a language model for the first time in the pill identification system and proposed a novel coordinate encoding technique for effective correction in the language model. We identified pills using similarity scores of pill characteristics with those in the database. We collected the open pill database from South Korea and the United States in May 2022. We used a total of 24,404 pill images in our experiments. The experimental results show that the predicted top-1 candidates achieve accuracy levels of 85.6% (South Korea) and 74.5% (United States) for the types of pills not trained on 2 different databases (South Korea and the United States). Furthermore, the predicted top-1 candidate accuracy of our system was 78% with consumer-granted images, which was achieved by training only 1 image per pill. The results demonstrate that our system could identify and retrieve new pills without additional model updates. Finally, we confirmed through an ablation study that the language model that we emphasized significantly improves the pill identification ability of the system. Our study proposes the possibility of reducing medical errors by showing that the introduction of artificial intelligence can identify numerous pills with high precision in real time. Our study suggests that the proposed system can reduce patients' misuse of medications and help medical staff focus on higher-level tasks by simplifying time-consuming lower-level tasks such as pill identification.

Machine Learning Estimation of the Phase at the Fading Points of an OFDR-Based Distributed Sensor.

The paper reports a machine learning approach for estimating the phase in a distributed acoustic sensor implemented using optical frequency domain reflectometry, with enhanced robustness at the fading points. A neural network configuration was trained using a simulated set of optical signals that were modeled after the Rayleigh scattering pattern of a perturbed fiber. Firstly, the performance of the network was verified using another set of numerically generated scattering profiles to compare the achieved accuracy levels with the standard homodyne detection method. Then, the proposed method was tested on real experimental measurements, which indicated a detection improvement of at least 5.1 dB with respect to the standard approach.

Study on Rice Grain Mildewed Region Recognition Based on Microscopic Computer Vision and YOLO-v5 Model.

This study aims to develop a high-speed and nondestructive mildewed rice grain detection method. First, a set of microscopic images of rice grains contaminated by Aspergillus niger, Penicillium citrinum, and Aspergillus cinerea are acquired to serve as samples, and the mildewed regions are marked. Then, three YOLO-v5 models for identifying regions of rice grain with contamination of Aspergillus niger, Penicillium citrinum, and Aspergillus cinerea in microscopic images are established. Finally, the relationship between the proportion of mildewed regions and the total number of colonies is analyzed. The results show that the proposed YOLO-v5 models achieve accuracy levels of 89.26%, 91.15%, and 90.19% when detecting mildewed regions with contamination of Aspergillus niger, Penicillium citrinum, and Aspergillus cinerea in the microscopic images of the verification set. The proportion of the mildewed region area of rice grain with contamination of Aspergillus niger/Penicillium citrinum/Aspergillus cinerea is logarithmically correlated with the logarithm of the total number of colonies (TVC). The corresponding determination coefficients are 0.7466, 0.7587, and 0.8148, respectively. This study provides a reference for future research on high-speed mildewed rice grain detection methods based on MCV technology.

Keeping Track of Hawaii's Bottomfish Populations With the Help of Citizen Scientists

The “Deep 7” bottomfish complex, which consists of six snapper and one grouper species, is a complex that carries high economic and cultural importance to the islands of Hawaii. These bottomfish have been monitored through the National Oceanic and Atmospheric Administration Pacific Islands Fisheries Science Center's Deep 7 fishery‐independent surveys since 2016. These surveys use underwater stereo camera systems that produce hundreds of thousands of images that must be annotated by human analysts in order to generate species‐specific, size‐structured abundance estimates. We developed a citizen science project, called “OceanEYEs,” as a means to effectively process this imagery. A beta test was conducted to determine the accuracy of citizen science annotations in comparison to expert annotators. Our results suggest that aggregated citizen scientist data can achieve accuracy levels approaching that of expert annotators, which has the potential to improve image annotation efficiency and produce large volumes of high‐quality training data to improve machine learning algorithms.

Combination of pre-treatment dynamic [18F]FET PET radiomics and conventional clinical parameters for the survival stratification in patients with IDH-wildtype glioblastoma

PurposeThe aim of this study was to build and evaluate a prediction model which incorporates clinical parameters and radiomic features extracted from static as well as dynamic [18F]FET PET for the survival stratification in patients with newly diagnosed IDH-wildtype glioblastoma.MethodsA total of 141 patients with newly diagnosed IDH-wildtype glioblastoma and dynamic [18F]FET PET prior to surgical intervention were included. Patients with a survival time ≤ 12 months were classified as short-term survivors. First order, shape, and texture radiomic features were extracted from pre-treatment static (tumor-to-background ratio; TBR) and dynamic (time-to-peak; TTP) images, respectively, and randomly divided into a training (n = 99) and a testing cohort (n = 42). After feature normalization, recursive feature elimination was applied for feature selection using 5-fold cross-validation on the training cohort, and a machine learning model was constructed to compare radiomic models and combined clinical-radiomic models with selected radiomic features and clinical parameters. The area under the ROC curve (AUC), accuracy, sensitivity, specificity, and positive and negative predictive values were calculated to assess the predictive performance for identifying short-term survivors in both the training and testing cohort.ResultsA combined clinical-radiomic model comprising six clinical parameters and six selected dynamic radiomic features achieved highest predictability of short-term survival with an AUC of 0.74 (95% confidence interval, 0.60–0.88) in the independent testing cohort.ConclusionsThis study successfully built and evaluated prediction models using [18F]FET PET-based radiomic features and clinical parameters for the individualized assessment of short-term survival in patients with a newly diagnosed IDH-wildtype glioblastoma. The combination of both clinical parameters and dynamic [18F]FET PET–based radiomic features reached highest accuracy in identifying patients at risk. Although the achieved accuracy level remained moderate, our data shows that the integration of dynamic [18F]FET PET radiomic data into clinical prediction models may improve patient stratification beyond established prognostic markers.