Train Machine Learning Models Research Articles

Codebase release 1.3 for QArray

Semiconductor quantum dot arrays are a leading architecture for the development of quantum technologies. Over the years, the constant capacitance model has served as a fundamental framework for simulating, understanding, and navigating the charge stability diagrams of small quantum dot arrays. However, while the size of the arrays keeps growing, solving the constant capacitance model becomes computationally prohibitive. This paper presents an open-source software package able to compute a 100 × 100100×100 pixels charge stability diagram of a 16-dot array in less than a second. Smaller arrays can be simulated in milliseconds - faster than they could be measured experimentally, enabling the creation of diverse datasets for training machine learning models and the creation of digital twins that can interface with quantum dot devices in real-time. Our software package implements its core functionalities in the systems programming language Rust and the high-performance numerical computing library JAX. The Rust implementation benefits from advanced optimisations and parallelisation, enabling the users to take full advantage of multi-core processors. The JAX implementation allows for GPU acceleration.

The digital fingerprint of learner behavior: Empirical evidence for individuality in learning using deep learning

Personalized learning builds upon the fundamental assumption of uniqueness in learning behavior, often taken for granted. Quite surprisingly, however, the literature provides little to no empirical evidence backing the existence of individual learning behaviors. Driven by curiosity, we challenge this axiom. Our operationalization of a unique learning behavior draws an analogy to a fingerprint – a distinctive trait that sets individuals apart, which we correspondingly termed the ‘Digital Fingerprint of Learner Behavior’ (DFL). If such a thing as DFL truly exists, then given enough fine-grained behavioral data, we argue that it should be possible to model a DFL to a level of discriminability that enables training machine learning models to associate (map) between the (de-identified) digital traces of the same learner in diverse contexts. To test our hypothesis, we experimented with data from 24 MITx massive open online courses (MOOCs) offered via edX between 2014 and 2017. We focused our investigation on contexts where both the content and platform remain constant. A learner's DFL was computed from the learner's activity data within a specific course chapter, as stored in the system's logs. The results show that the mean level of accuracy (across courses) in identifying unseen DFLs is 0.582 (SD=0.173). Using Shapley Additive exPlanations (SHAP), we rank 686 features for their importance in differentiating between DFLs. To the best of our knowledge, this study is the first to provide empirical evidence that learners' behavior is unique to a degree that can distinguish between them on an individual level, similar to the level of identification provided by a fingerprint, and sets a benchmark for the task of DFL identification.

FedNIC: enhancing privacy-preserving federated learning via homomorphic encryption offload on SmartNIC

Federated learning (FL) has emerged as a promising paradigm for secure distributed machine learning model training across multiple clients or devices, enabling model training without having to share data across the clients. However, recent studies revealed that FL could be vulnerable to data leakage and reconstruction attacks even if the data itself are never shared with another client. Thus, to resolve such vulnerability and improve the privacy of all clients, a class of techniques, called privacy-preserving FL, incorporates encryption techniques, such as homomorphic encryption (HE), to encrypt and fully protect model information from being exposed to other parties. A downside to this approach is that encryption schemes like HE are very compute-intensive, often causing inefficient and excessive use of client CPU resources that can be used for other uses. To alleviate this issue, this study introduces a novel approach by leveraging smart network interface cards (SmartNICs) to offload compute-intensive HE operations of privacy-preserving FL. By employing SmartNICs as hardware accelerators, we enable efficient computation of HE while saving CPU cycles and other server resources for more critical tasks. In addition, by offloading encryption from the host to another device, the details of encryption remain secure even if the host is compromised, ultimately improving the security of the entire FL system. Given such benefits, this paper presents an FL system named FedNIC that implements the above approach, with an in-depth description of the architecture, implementation, and performance evaluations. Our experimental results demonstrate a more secure FL system with no loss in model accuracy and up to 25% in reduced host CPU cycle, but with a roughly 46% increase in total training time, showing the feasibility and tradeoffs of utilizing SmartNICs as an encryption offload device in federated learning scenarios. Finally, we illustrate promising future study and potential optimizations for a more secure and privacy-preserving federated learning system.

Phylodynamic Structure in the Botswana HIV Epidemic.

Background Studying viral sequences can provide insights into the structure of host contact networks through which the virus is transmitted. Uncovering the population structure of the HIV-1 epidemic in Botswana will help optimise public health interventions and may identify hidden sub-epidemics. We sought to determine the phylodynamic structure of the Botswana HIV-1 epidemic from viral sequence genetic data. Methods The Botswana Combination Prevention Project (BCPP) randomly sampled 20% of households in 30 villages in Botswana between 2013-2018 and tested for HIV-1. Extensive demographic data were collected from all participants and next-generation full-genome HIV-1 sequences were generated from HIV-1 positive participants (n = 4,164), 78% of whom were on antiretroviral treatment (ART). We inferred the stage of infection (< or > 1 year) among HIV-1 cases based on nucleotide diversity and clinical data using a previously trained machine learning model. We then reconstructed time-resolved gag and pol phylogenies from sequences, other Botswana cohorts and publicly available sequences that were genetically close to those from Botswana. We statistically explored phylogenies for partitions with diverging patterns of coalescence, indicating sub-epidemics, and estimated viral effective population size through time, a measure of viral incidence, for each partition. Finally, we compared the demographic makeup, clinical and geographic characteristics across partitions using χ2, ANOVA tests and Tukey analysis. Results We identified three partitions of time-resolved gag and pol phylogenies, revealing divergent patterns of coalescence and HIV-1 transmission. In both gag and pol phylogenies, partitions with persistent growth and transmission were characterised by lower treatment coverage and more recent infections when compared to other partitions. The Southern and South East regions of Botswana were over-represented in the fast-growing partitions. Conclusion Our findings suggest that transmission is slowing in segments of the population that have high ART coverage. However, recent infections are over-represented in ongoing sub-epidemics. The phylodynamic structure suggests that there are districts with higher growth and prioritising these in the deployment of public health interventions might curb new infections. Nonetheless the high mobility of Botswana residents should be taken into consideration in implementing effective interventions to combat HIV-1.

Search for Young Stellar Objects within 4XMM-DR13 Using CatBoost and SPE

Classifying and summarizing large data sets from different sky survey projects is essential for various subsequent scientific research. By combining data from 4XMM-DR13, Sloan Digital Sky Survey (SDSS) DR18, and CatWISE, we formed an XMM-WISE-SDSS sample that included information in the X-ray, optical, and infrared bands. By cross matching this sample with data sets from known spectral classifications from SDSS and LAMOST, we obtained a training data set containing stars, galaxies, quasars, and young stellar objects (YSOs). Two machine learning methods, CatBoost and Self-Paced Ensemble (SPE), were used to train and construct machine learning models through training sets to classify the XMM-WISE-SDSS sample. Notably, the SPE classifier showed excellent performance in YSO classification, identifying 1102 YSO candidates from 160,545 sources, including 258 known YSOs. Then we further verify whether these candidates are YSOs by the spectra in LAMOST and the identification in the SIMBAD and VizieR databases. Finally there are 412 unidentified YSO candidates. The discovery of these new YSOs is an important addition to existing YSO samples and will deepen our understanding of star formation and evolution. Moreover we provided a classification catalog for the whole XMM-WISE-SDSS sample.

Blockchain-Based Federated Learning: A Survey and New Perspectives

Federated learning, as a novel distributed machine learning mode, enables the training of machine learning models on multiple devices while ensuring data privacy. However, the existence of single-point-of-failure bottlenecks, malicious threats, scalability of federated learning implementation, and lack of incentive mechanisms have seriously hindered the development of federated learning technology. In recent years, as a distributed ledger, blockchain has the characteristics of decentralization, tamper-proof, transparency, security, etc., which can solve the issues encountered in the above-mentioned federated learning. Particularly, the integration of federated learning and blockchain leads to a new paradigm, called blockchain-based federated learning (BFL), which has been successfully applied in many application scenarios. This paper aims to provide a comprehensive review of recent efforts on blockchain-based federated learning. More concretely, we propose and design a taxonomy of blockchain-based federated learning models, along with providing a comprehensive summary of the state of the art. Various applications of federated learning based on blockchain are introduced. Finally, we expand on current trends and provide new perspectives pertaining to this new and exciting development in the field.

Artificial intelligence for predicting arctic permafrost and active layer temperatures along the Alaskan North Slope

AbstractClimate change in Arctic regions poses a threat to permafrost stability, potentially leading to issues such as infrastructure damage, altered ecosystems, and increased greenhouse gas emissions. The challenge of monitoring permafrost temperatures and identifying critical environmental factors is accentuated by the scarcity of subsurface thermal data in remote Arctic locales. To alleviate this lack of soil thermal data we set out to combine in situ observed soil temperatures with MERRA-2 reanalysis data and Machine Learning (ML) to develop local and season-specific subsurface soil temperature predictions in Alaska. First, reanalysis features are selected based on surface energy budget physics. Coupled with Julian calendar dates, reanalysis features are preprocessed and then fed to the ML prediction model. We conducted a series of computational experiments and tested the results against performance benchmarks to determine the optimal prediction models, length of look-back period of model inputs, and the training set size. Six conventional ML models (e.g., GBDT, RF, SVR) and five statistical baselines (e.g. ARIMA) using in situ time series of soil temperature data ranging from 0 - 1 m depth are considered for the prediction task. The models are trained using in situ soil temperatures at depths between 0 and 1 m, spanning 16+ years, from field sites at Deadhorse and Toolik Lake, Alaska. All prediction performances are assessed using root mean squared error (RMSE) and mean absolute error (MAE). Results show that locally trained ML models can estimate shallow soil temperatures with an average error of $$RMSE=1.308^{\circ }$$ R M S E = 1 . 308 ∘ C.

Prediction of Attention Groups and Big Five Personality Traits from Gaze Features Collected from an Outlier Search Game.

This study explores the intersection of personality, attention and task performance in traditional 2D and immersive virtual reality (VR) environments. A visual search task was developed that required participants to find anomalous images embedded in normal background images in 3D space. Experiments were conducted with 30 subjects who performed the task in 2D and VR environments while their eye movements were tracked. Following an exploratory correlation analysis, we applied machine learning techniques to investigate the predictive power of gaze features on human data derived from different data collection methods. Our proposed methodology consists of a pipeline of steps for extracting fixation and saccade features from raw gaze data and training machine learning models to classify the Big Five personality traits and attention-related processing speed/accuracy levels computed from the Group Bourdon test. The models achieved above-chance predictive performance in both 2D and VR settings despite visually complex 3D stimuli. We also explored further relationships between task performance, personality traits and attention characteristics.

CODI: Enhancing machine learning-based molecular profiling through contextual out-of-distribution integration

Abstract Molecular analytics increasingly utilize machine learning (ML) for predictive modeling based on data acquired through molecular profiling technologies. However, developing robust models that accurately capture physiological phenotypes is challenged by the dynamics inherent to biological systems, variability stemming from analytical procedures, and the resource-intensive nature of obtaining sufficiently representative datasets. Here, we propose and evaluate a new method: Contextual Out-of-Distribution Integration (CODI). Based on experimental observations, CODI generates synthetic data that integrate unrepresented sources of variation encountered in real-world applications into a given molecular fingerprint dataset. By augmenting a dataset with out-of-distribution variance, CODI enables an ML model to better generalize to samples beyond the seed training data, reducing the need for extensive experimental data collection. Using three independent longitudinal clinical studies and a case-control study, we demonstrate CODI's application to several classification tasks involving vibrational spectroscopy of human blood. We showcase our approach's ability to enable personalized fingerprinting for multi-year longitudinal molecular monitoring and enhance the robustness of trained ML models for improved disease detection. Our comparative analyses reveal that incorporating CODI into the classification workflow consistently leads to increased robustness against data variability and improved predictive accuracy.

Prediction of seismic performance of steel frame structures: A machine learning approach

The study of seismic performance in structural engineering is deemed crucial due to the intricate and unpredictable nature of earthquakes. This study explores advanced seismic performance prediction in structural engineering using machine learning techniques. Non-Linear Dynamic Analysis (NDA) was conducted on Steel Moment-Resisting Frames (SMRFs) situated on soil type D in seismic zone II with varying configurations using ETABS software. A substantial dataset comprising 29,200 data points from 292 models was generated to train machine learning models aimed at predicting the Maximum Inter-Story Drift Ratio (M-IDR), a critical parameter for assessing seismic limit-state capacity. The machine learning models, including Random Forest (RF), Extreme Gradient Boosting Machine (XGBoost), and Artificial Neural Networks (ANN), demonstrated high accuracy with R2 of 0.9625, 0.95327, and 0.94247 respectively, indicating a robust correlation between predicted and actual values. These results imply that the trained models can effectively predict seismic performance with high precision. A user-friendly Graphical User Interface (GUI) was developed using the trained models to facilitate the practical application of these models, significantly reducing computational costs and analytical efforts for researchers and engineers. The findings underscore the potential of integrating machine learning with structural engineering to enhance seismic performance predictions, contributing to the development of safer and more resilient structures.

Machine learning applications of network security enhancement: review

Machine learning (ML) is being used to improve intrusion detection mechanisms and identification in cyber security. Network data volume scaling (with the help of Machine learning) — Automated analysis and pattern recognition for large amounts of network-data, thereby detection of anomalies / potentially malicious activities that escape current rule-based techniques. By training ML models on historical data, these models can learn benign network behavior as well the anomalies in them that may result from malicious activities. The purpose of this essay/report is to begin taking a look under the hood at how ML can be used for security threat detection and analysis in-networking on an ongoing basis. This paper covers the automation of ML algorithms to enhance network security. Given the current state of electronic threats and their evolution, traditional security methods are typically insufficient. ML can analyze large volumes of data, learn patterns from it and this makes it suitable to complement network defense mechanisms. It then discusses different ML applications in network cybersecurity: intrusion detection, anomaly detection, spam and malware analysis; which the paper characterizes. It analyzes the potential, benefits and constrains of major ML methods in network security like supervised learning; unsupervised learning and reinforcement-learning. Finally, this paper represents recent progress in the use and impact of ML techniques along with case studies.The paper discusses the existing difficulties in the field such as the necessity for datasets and the vulnerability of machine learning models to adversarial attacks.T he paper also highlights avenues for exploration by focusing on developing scalable security solutions based on machine learning that are resilient and flexible.The goal of this examination is to offer both researchers and industry professionals valuable perspectives into the opportunities and obstacles linked to utilizing machine learning, in the domain of network security. ML methods have potential to improve network security by addressing the challenges posed by the increasing cyber threats that traditional security measures struggle to combat effectively over time. One key strength of ML lies in its capacity to analyze datasets and identify intricate patterns efficiently. The research paper delves into applications of ML in enhancing network security. The list covers security tools like Intrusion Detection Systems (IDS) examining malware and phishing attempts as well as anomalies in network activity and user behavior analysis (UEBA). The study explores both supervised and unsupervised learning methods. How they are used for quick threat detection and response in real time scenarios.You will find case studies and recent developments that showcase the implementation and effectiveness of these strategies.In addition the article delves into the obstacles linked to using machine learning techniques in network security including the necessity, for datasets, The paper's goal is to give an enlightening summary to scholars and practitioners about how machine learning can be applied to network security in order to provide solutions that are robust, adaptive, and scalable. To this end, it touches on several relevant aspects. One is the threat posed by adversarial attacks on the sorts of models that are likely to be used in this context. Another is the imperative, deriving from both adversarial threat and model drift, that models needed in this context be available in a form usable for continuous update. Keywords: Network Security, machine learning (ML), Intrusion Detection Systems (IDS), entity behavior analytics (UEBA).

Recovering MRI magnetic field strength properties using machine learning based on image compression quality scores

AbstractOver the years, significant hardware advancements have led to higher magnetic field MRI (Magnetic Resonance Imaging) acquisitions, providing improved diagnostic image quality at a higher cost. While higher image quality is desired, the investment required for high-field MRI imaging systems remains a significant consideration. Furthermore, historical patient records may still contain low-field magnetic strength MRI data. In this work, we present how image compression quality scores of MRI images could be used to recover their magnetic field strength properties. This work presents an interesting inverse problem scenario where output properties are used to recover an input property of an image. We implemented supervised machine learning models that utilize compressed image quality scores - based on Mean Squared Error (MSE), Structural Similarity Index (SSIM), Visual Information Fidelity (VIF), and Earth Movers Distance (EM) metrics - as inputs. These models classify images into specific classes of magnetic field strength at which they were acquired (Low, Medium, and High) based solely on the quality scores of their compressed copies. The trained machine learning models (Support Vector Machine (SVM), Logistic Regression, K-nearest Neighbours (KNN), Decision Tree and Random Forest) achieved nearly 100% performance efficiency based on accuracy and F-1 score for all considered quality measures. This comprehensive evaluation offers insights into how compression techniques and quality factors impact image fidelity across various MRI magnetic field strengths. The findings of this study may prove valuable in recovering missing meta-tag information in historical image DICOM records and Context-Based Retrieval Systems (CBRS). The approach of using quality scores as features in training machine learning models, as demonstrated in this work, can be extended to other image groups where input parameters create statistically significant distinctions or are challenging to extract from images.

Machine learning-based outlier detection for pipeline in-line inspection data

Pipeline companies are facing challenges in maintaining the integrity and reliability of their pipelines. They are working towards predictive maintenance using machine learning-based approaches to predicting anomalies. Training machine learning models requires sufficient data. Data quality is therefore becoming important because inaccurate data will lead to an inaccurate or wrong decision on pipeline condition assessment and the following management. This research paper intends to address the data quality issues of pipeline inspection data such as in-line inspection (ILI) data using machine learning models. Different machine learning models developed by random forest regression, linear regression, and nearest neighbors’ methods were tested to detect outliers in the ILI data. In this paper, the ILI data collected from an oil pipeline over a period of 22 years was applied to testing and analysis. To verify the outlier detection results of machine learning models, we used statistical analysis including Z-score method to check and find if there are any gaps in the analysis. It verifies that all these methods show almost the same or very similar results for the detection of the outliers. Hence, this study presents a robust method for the field applications in the pipeline industry.

Machine Learning Algorithm for Enhanced Cybersecurity: Identification and Mitigation of Emerging Threats

Machine learning (ML) methodologies have significantly transformed cybersecurity by offering sophisticated instruments that not only identify but also avert and alleviate cyber threats. This research study seeks to examine the convergence of machine learning and cybersecurity, focusing on diverse methodologies and their use in enhancing cybersecurity measures. The study examines several machines learning methods, including Graph Neural Networks, Adversarial Learning, Federated Learning, Explainable AI, and Reinforcement Learning. Every algorithm is essential for enhancing the identification and mitigation of cyber assaults. Graph Neural Networks facilitate the modelling of intricate linkages within cybersecurity data. It aids not just in forecasting future events but also in identifying anomalies and analyzing network traffic. Adversarial Learning assists in training machine learning models to address the difficulty of producing misleading input data that can deceive any model, hence enhancing their efficacy. Federated Learning is examined as a method for training machine learning models across decentralized networks while preserving data privacy and enhancing model accuracy. Explainable AI methodologies primarily offer transparency and interpretability in machine learning-driven cybersecurity decisions, which are crucial for comprehension and confidence in automated security systems. Reinforcement Learning is focused on a trial-and-error methodology, wherein the model acquires new tasks through a system of rewards and penalties. These sophisticated algorithms jointly improve the effectiveness, precision, and clarity of cybersecurity protocols, offering strong protection against emerging cyber threats.

Phase Stability of CH4 and CO2 Hydrates under Confinement Predicted by Machine Learning.

Understanding the phase stability of gas hydrates under confinement is fundamental to the geological stability evolutions of gas hydrate systems on Earth. Herein, the phase stability of CH4 and CO2 hydrates under confinement is predicted by machine learning. Three machine learning models, including support vector machine, random forest, and gradient boosting decision tree, are constructed to predict the phase stability of CH4 and CO2 hydrates under confinement. Our machine learning results show that the prediction accuracy of the support vector machine model is highest, yet the prediction accuracy of the random forest model is lowest among those machine learning models in determining the phase stability of confined gas hydrates. Based on their performance in predicting the phase stability of confined gas hydrates, the support vector machine model with a training set fraction of 0.7 is finally chosen to deal with the unknown phase stability of confined gas hydrates. Importantly, the average accuracy of the support vector machine model can reach more than 90% in predicting the unknown phase stability of both CH4 and CO2 hydrates. The trained machine learning models can help us to quickly and accurately determine the phase stability of CH4 and CO2 hydrates under confinement in future applications.

ProstateZones – Segmentations of the prostatic zones and urethra for the PROSTATEx dataset

Manual segmentations are considered the gold standard for ground truth in machine learning applications. Such tasks are tedious and time-consuming, albeit necessary to train reliable models. In this work, we present a dataset with expert segmentations of the prostatic zones and urethra for 200 randomly selected patients from the PROSTATEx dataset. Notably, independent duplicate segmentations were performed for 40 patients, providing inter-reader variability data. This results in a total of 240 segmentations. This dataset can be used to train machine learning models or serve as an external test set for evaluating models trained on private data, thereby addressing a current gap in the field. The delineated structures and terminology adhere to the latest Prostate Imaging Reporting and Data Systems v2.1 guidelines, ensuring consistency.

Survey on Federated Learning for Intrusion Detection System: Concept, Architectures, Aggregation Strategies, Challenges, and Future Directions

Intrusion Detection Systems (IDS) are essential for securing computer networks by identifying and mitigating potential threats. However, traditional IDS face challenges related to scalability, privacy, and computational demands as network data complexity increases. Federated Learning (FL) has emerged as a promising solution, enabling collaborative model training on decentralized data sources while preserving data privacy. Each participant retains local data repositories, ensuring data sovereignty and precluding data sharing. Leveraging the FL framework, participants locally train machine learning models on their respective datasets, subsequently transmitting model updates to a central server for aggregation. The central server then disseminates the aggregated model updates to individual participants, collectively striving to bolster intrusion detection capabilities. This article presents a comprehensive survey of FL applications in IDS, covering core concepts, architectural approaches, and aggregation strategies. We evaluate the strengths and limitations of various FL methodologies for IDS, addressing privacy and security concerns and exploring privacy-preserving techniques and security protocols. Our examination of aggregation strategies within the FL framework for IDS aims to highlight their effectiveness, limitations, and potential enhancements.

Dissecting the Reduced Penetrance of Putative Loss-of-Function Variants in Population-Scale Biobanks.

Loss-of-function variants (LoFs) disrupt the activity of their impacted gene. They are often associated with clinical phenotypes, including autosomal dominant diseases driven by haploinsufficiency. Recent analyses using biobanks have suggested that LoF penetrance for some haploinsufficient disorders may be low, an observation that has important implications for population genomic screening. However, biobanks are also rife with missing data, and the reliability of these findings remains uncertain. Here, we examine the penetrance of putative LoFs (pLoFs) using a cohort of ≈24,000 carriers derived from two population-scale biobanks: the UK Biobank and the All of Us Research Program. We investigate several possible etiologies for reduced pLoF penetrance, including biobank recruitment biases, annotation artifacts, missed diagnoses, and incomplete clinical records. Systematically accounting for these factors increased penetrance, but widespread reduced penetrance remained. Therefore, we hypothesized that other factors must be driving this phenomenon. To test this, we trained machine learning models to identify pLoFs with high penetrance using the genomic features specific to each variant. These models were predictive of penetrance across a range of diseases and ploF types, including those with prior evidence for pathogenicity. This suggests that reduced ploF penetrance is in fact common, and care should be taken when counseling asymptomatic carriers.

Predicting the Malignancy Grade of Soft Tissue Sarcomas on MRI Using Conventional Image Reading and Radiomics

Objectives: This study aims to investigate MRI features predicting the grade of STS malignancy using conventional image reading and radiomics. Methods: Pretherapeutic imaging data regarding size, tissue heterogeneity, peritumoral changes, necrosis, hemorrhage, and cystic degeneration were evaluated in conventional image reading. Furthermore, the tumors’ apparent diffusion coefficient (ADC) values and radiomics features were extracted and analyzed. A random forest machine learning algorithm was trained and evaluated based on the extracted features. Results: A total of 139 STS cases were included in this study. The mean tumor ADC and the ratio between tumor ADC to healthy muscle ADC were significantly lower in high-grade tumors (p = 0.001 and 0.005, respectively). Peritumoral edema (p &lt; 0.001) and peritumoral contrast enhancement (p &lt; 0.001) were significantly more extensive in high-grade tumors. Tumor heterogeneity was significantly increased in high-grade sarcomas, particularly in T2w- and contrast-enhanced sequences using conventional image reading (p &lt; 0.001) as well as in the radiomics analysis (p &lt; 0.001). Our trained random forest machine learning model predicted high-grade status with an area under the curve (AUC) of 0.97 and an F1 score of 0.93. Biopsy-underestimated tumors exhibited differences in tumor heterogeneity and peritumoral changes. Conclusions: Tumor heterogeneity is a key characteristic of high-grade STSs, which is discernible through conventional imaging reading and radiomics analysis. Higher STS grades are also associated with low ADC values, peritumoral edema, and peritumoral contrast enhancement.

Digital voice analysis as a biomarker of acromegaly.

There is a considerable diagnostic delay in acromegaly contributing to increased morbidity. Voice changes due to orofacial and laryngeal changes are common in acromegaly. Our aim was to explore the use of digital voice analysis as a biomarker for acromegaly using broad acoustic analysis and machine learning. Voice recordings from patients with acromegaly and matched controls were collected using a mobile phone at Swedish university hospitals. Anthropometric and clinical data and the Voice Handicap Index (VHI) were assessed. Digital voice analysis of a sustained and stable vowel [a] resulted in 3274 parameters, which were used for training of machine learning models classifying the speaker as "acromegaly" or "control". The machine learning model was trained with 76% of the data and the remaining 24% was used to assess its performance. For comparison, voice recordings of 50 pairs of participants were assessed by 12 experienced endocrinologists. We included 151 Swedish patients with acromegaly (13% biochemically active and 10% newly diagnosed) and 139 matched controls. The machine learning model identified patients with acromegaly more accurately [area under the receiver operating curve (ROC AUC) 0.84] than experienced endocrinologists (ROC AUC 0.69). Self-reported voice problems were more pronounced in patients with acromegaly than matched controls (median VHI 6 vs 2, P < .01) with higher prevalence of clinically significant voice handicap (VHI ≥20: 22.5% vs 3.6%). Digital voice analysis can identify patients with acromegaly from short voice recordings with high accuracy. Patients with acromegaly experience more voice disorders than matched controls.