Deep learning-enabled medical computer vision

Artificial intelligence model for tumoral clinical decision support systems

BACKGROUND AND OBJECTIVE

Comparative diagnostic in brain tumor evaluation makes possible to use the available information of a medical center to compare similar cases when a new patient is evaluated. By leveraging Artificial Intelligence models, the proposed system is able of retrieving the most similar cases of brain tumors for a given query. The primary objective is to enhance the diagnostic process by generating more accurate representations of medical images, with a particular focus on patient-specific normal features and pathologies. A key distinction from previous models lies in its ability to produce enriched image descriptors solely from binary information, eliminating the need for costly and difficult to obtain tumor segmentation.

METHODS

The proposed model uses Artificial Intelligence to detect patient features to recommend the most similar cases from a database. The system not only suggests similar cases but also balances the representation of healthy and abnormal features in its design. This not only encourages the generalization of its use but also aids clinicians in their decision-making processes. This generalization makes possible for future research in different medical diagnosis areas with almost not any change in the system.

RESULTS

We conducted a comparative analysis of our approach in relation to similar studies. The proposed architecture obtains a Dice coefficient of 0.474 in both tumoral and healthy regions of the patients, which outperforms previous literature. Our proposed model excels at extracting and combining anatomical and pathological features from brain Magnetic Resonances (MRs), achieving state-of-the-art results while relying on less expensive label information. This substantially reduces the overall cost of the training process. Our findings highlight the significant potential for improving the efficiency and accuracy of comparative diagnostics and the treatment of tumoral pathologies.

CONCLUSIONS

This paper provides substantial grounds for further exploration of the broader applicability and optimization of the proposed architecture to enhance clinical decision-making. The novel approach presented in this work marks a significant advancement in the field of medical diagnosis, particularly in the context of Artificial Intelligence-assisted image retrieval, and promises to reduce costs and improve the quality of patient care using Artificial Intelligence as a support tool instead of a black box system.

W-DRAG: A joint framework of WGAN with data random augmentation optimized for generative networks for bone marrow edema detection in dual energy CT

Dual-energy computed tomography (CT) is an excellent substitute for identifying bone marrow edema in magnetic resonance imaging. However, it is rarely used in practice owing to its low contrast. To overcome this problem, we constructed a framework based on deep learning techniques to screen for diseases using axial bone images and to identify the local positions of bone lesions. To address the limited availability of labeled samples, we developed a new generative adversarial network (GAN) that extends expressions beyond conventional augmentation (CA) methods based on geometric transformations. We theoretically and experimentally determined that combining the concepts of data augmentation optimized for GAN training (DAG) and Wasserstein GAN yields a considerably stable generation of synthetic images and effectively aligns their distribution with that of real images, thereby achieving a high degree of similarity. The classification model was trained using real and synthetic samples. Consequently, the GAN technique used in the diagnostic test had an improved F1 score of approximately 7.8% compared with CA. The final F1 score was 80.24%, and the recall and precision were 84.3% and 88.7%, respectively. The results obtained using the augmented samples outperformed those obtained using pure real samples without augmentation. In addition, we adopted explainable AI techniques that leverage a class activation map (CAM) and principal component analysis to facilitate visual analysis of the network's results. The framework was designed to suggest an attention map and scattering plot to visually explain the disease predictions of the network.

Validation and application of computer vision algorithms for video-based tremor analysis

Tremor is one of the most common neurological symptoms. Its clinical and neurobiological complexity necessitates novel approaches for granular phenotyping. Instrumented neurophysiological analyses have proven useful, but are highly resource-intensive and lack broad accessibility. In contrast, bedside scores are simple to administer, but lack the granularity to capture subtle but relevant tremor features. We utilise the open-source computer vision pose tracking algorithm Mediapipe to track hands in clinical video recordings and use the resulting time series to compute canonical tremor features. This approach is compared to marker-based 3D motion capture, wrist-worn accelerometry, clinical scoring and a second, specifically trained tremor-specific algorithm in two independent clinical cohorts. These cohorts consisted of 66 patients diagnosed with essential tremor, assessed in different task conditions and states of deep brain stimulation therapy. We find that Mediapipe-derived tremor metrics exhibit high convergent clinical validity to scores (Spearman's ρ = 0.55-0.86, p≤ .01) as well as an accuracy of up to 2.60 mm (95% CI [-3.13, 8.23]) and ≤0.21 Hz (95% CI [-0.05, 0.46]) for tremor amplitude and frequency measurements, matching gold-standard equipment. Mediapipe, but not the disease-specific algorithm, was capable of analysing videos involving complex configurational changes of the hands. Moreover, it enabled the extraction of tremor features with diagnostic and prognostic relevance, a dimension which conventional tremor scores were unable to provide. Collectively, this demonstrates that current computer vision algorithms can be transformed into an accurate and highly accessible tool for video-based tremor analysis, yielding comparable results to gold standard tremor recordings.

Development and validation of a smartphone-based deep-learning-enabled system to detect middle-ear conditions in otoscopic images

Middle-ear conditions are common causes of primary care visits, hearing impairment, and inappropriate antibiotic use. Deep learning (DL) may assist clinicians in interpreting otoscopic images. This study included patients over 5 years old from an ambulatory ENT practice in Strasbourg, France, between 2013 and 2020. Digital otoscopic images were obtained using a smartphone-attached otoscope (Smart Scope, Karl Storz, Germany) and labeled by a senior ENT specialist across 11 diagnostic classes (reference standard). An Inception-v2 DL model was trained using 41,664 otoscopic images, and its diagnostic accuracy was evaluated by calculating class-specific estimates of sensitivity and specificity. The model was then incorporated into a smartphone app called i-Nside. The DL model was evaluated on a validation set of 3,962 images and a held-out test set comprising 326 images. On the validation set, all class-specific estimates of sensitivity and specificity exceeded 98%. On the test set, the DL model achieved a sensitivity of 99.0% (95% confidence interval: 94.5-100) and a specificity of 95.2% (91.5-97.6) for the binary classification of normal vs. abnormal images; wax plugs were detected with a sensitivity of 100% (94.6-100) and specificity of 97.7% (95.0-99.1); other class-specific estimates of sensitivity and specificity ranged from 33.3% to 92.3% and 96.0% to 100%, respectively. We present an end-to-end DL-enabled system able to achieve expert-level diagnostic accuracy for identifying normal tympanic aspects and wax plugs within digital otoscopic images. However, the system's performance varied for other middle-ear conditions. Further prospective validation is necessary before wider clinical deployment.

Head movement dynamics in dystonia: a multi-centre retrospective study using visual perceptive deep learning

Dystonia is a neurological movement disorder characterised by abnormal involuntary movements and postures, particularly affecting the head and neck. However, current clinical assessment methods for dystonia rely on simplified rating scales which lack the ability to capture the intricate spatiotemporal features of dystonic phenomena, hindering clinical management and limiting understanding of the underlying neurobiology. To address this, we developed a visual perceptive deep learning framework that utilizes standard clinical videos to comprehensively evaluate and quantify disease states and the impact of therapeutic interventions, specifically deep brain stimulation. This framework overcomes the limitations of traditional rating scales and offers an efficient and accurate method that is rater-independent for evaluating and monitoring dystonia patients. To evaluate the framework, we leveraged semi-standardized clinical video data collected in three retrospective, longitudinal cohort studies across seven academic centres. We extracted static head angle excursions for clinical validation and derived kinematic variables reflecting naturalistic head dynamics to predict dystonia severity, subtype, and neuromodulation effects. The framework was also applied to a fully independent cohort of generalised dystonia patients for comparison between dystonia sub-types. Computer vision-derived measurements of head angle excursions showed a strong correlation with clinically assigned scores. Across comparisons, we identified consistent kinematic features from full video assessments encoding information critical to disease severity, subtype, and effects of neural circuit interventions, independent of static head angle deviations used in scoring. Our visual perceptive machine learning framework reveals kinematic pathosignatures of dystonia, potentially augmenting clinical management, facilitating scientific translation, and informing personalized precision neurology approaches.

Universal and extensible language-vision models for organ segmentation and tumor detection from abdominal computed tomography

The advancement of artificial intelligence (AI) for organ segmentation and tumor detection is propelled by the growing availability of computed tomography (CT) datasets with detailed, per-voxel annotations. However, these AI models often struggle with flexibility for partially annotated datasets and extensibility for new classes due to limitations in the one-hot encoding, architectural design, and learning scheme. To overcome these limitations, we propose a universal, extensible framework enabling a single model, termed Universal Model, to deal with multiple public datasets and adapt to new classes (e.g., organs/tumors). Firstly, we introduce a novel language-driven parameter generator that leverages language embeddings from large language models, enriching semantic encoding compared with one-hot encoding. Secondly, the conventional output layers are replaced with lightweight, class-specific heads, allowing Universal Model to simultaneously segment 25 organs and six types of tumors and ease the addition of new classes. We train our Universal Model on 3410 CT volumes assembled from 14 publicly available datasets and then test it on 6173 CT volumes from four external datasets. Universal Model achieves first place on six CT tasks in the Medical Segmentation Decathlon (MSD) public leaderboard and leading performance on the Beyond The Cranial Vault (BTCV) dataset. In summary, Universal Model exhibits remarkable computational efficiency (6× faster than other dataset-specific models), demonstrates strong generalization across different hospitals, transfers well to numerous downstream tasks, and more importantly, facilitates the extensibility to new classes while alleviating the catastrophic forgetting of previously learned classes. Codes, models, and datasets are available at https://github.com/ljwztc/CLIP-Driven-Universal-Model.

Artificial Intelligence in the Provision of Health Care: An American College of Physicians Policy Position Paper

Internal medicine physicians are increasingly interacting with systems that implement artificial intelligence (AI) and machine learning (ML) technologies. Some physicians and health care systems are even developing their own AI models, both within and outside of electronic health record (EHR) systems. These technologies have various applications throughout the provision of health care, such as clinical documentation, diagnostic image processing, and clinical decision support. With the growing availability of vast amounts of patient data and unprecedented levels of clinician burnout, the proliferation of these technologies is cautiously welcomed by some physicians. Others think it presents challenges to the patient-physician relationship and the professional integrity of physicians. These dispositions are understandable, given the "black box" nature of some AI models, for which specifications and development methods can be closely guarded or proprietary, along with the relative lagging or absence of appropriate regulatory scrutiny and validation. This American College of Physicians (ACP) position paper describes the College's foundational positions and recommendations regarding the use of AI- and ML-enabled tools and systems in the provision of health care. Many of the College's positions and recommendations, such as those related to patient-centeredness, privacy, and transparency, are founded on principles in the ACP Ethics Manual. They are also derived from considerations for the clinical safety and effectiveness of the tools as well as their potential consequences regarding health disparities. The College calls for more research on the clinical and ethical implications of these technologies and their effects on patient health and well-being.

Deep Learning for Automated Detection and Localization of Traumatic Abdominal Solid Organ Injuries on CT Scans

Computed tomography (CT) is the most commonly used diagnostic modality for blunt abdominal trauma (BAT), significantly influencing management approaches. Deep learning models (DLMs) have shown great promise in enhancing various aspects of clinical practice. There is limited literature available on the use of DLMs specifically for trauma image evaluation. In this study, we developed a DLM aimed at detecting solid organ injuries to assist medical professionals in rapidly identifying life-threatening injuries. The study enrolled patients from a single trauma center who received abdominal CT scans between 2008 and 2017. Patients with spleen, liver, or kidney injury were categorized as the solid organ injury group, while others were considered negative cases. Only images acquired from the trauma center were enrolled. A subset of images acquired in the last year was designated as the test set, and the remaining images were utilized to train and validate the detection models. The performance of each model was assessed using metrics such as the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value based on the best Youden index operating point. The study developed the models using 1302 (87%) scans for training and tested them on 194 (13%) scans. The spleen injury model demonstrated an accuracy of 0.938 and a specificity of 0.952. The accuracy and specificity of the liver injury model were reported as 0.820 and 0.847, respectively. The kidney injury model showed an accuracy of 0.959 and a specificity of 0.989. We developed a DLM that can automate the detection of solid organ injuries by abdominal CT scans with acceptable diagnostic accuracy. It cannot replace the role of clinicians, but we can expect it to be a potential tool to accelerate the process of therapeutic decisions for trauma care.

MA-MIL: Sampling point-level abnormal ECG location method via weakly supervised learning

BACKGROUND AND OBJECTIVE

Current automatic electrocardiogram (ECG) diagnostic systems could provide classification outcomes but often lack explanations for these results. This limitation hampers their application in clinical diagnoses. Previous supervised learning could not highlight abnormal segmentation output accurately enough for clinical application without manual labeling of large ECG datasets.

METHOD

In this study, we present a multi-instance learning framework called MA-MIL, which has designed a multi-layer and multi-instance structure that is aggregated step by step at different scales. We evaluated our method using the public MIT-BIH dataset and our private dataset.

RESULTS

The results show that our model performed well in both ECG classification output and heartbeat level, sub-heartbeat level abnormal segment detection, with accuracy and F1 scores of 0.987 and 0.986 for ECG classification and 0.968 and 0.949 for heartbeat level abnormal detection, respectively. Compared to visualization methods, the IoU values of MA-MIL improved by at least 17 % and at most 31 % across all categories.

CONCLUSIONS

MA-MIL could accurately locate the abnormal ECG segment, offering more trustworthy results for clinical application.

Maximizing Large Language Model Utility in Cardiovascular Care: A Practical Guide

Large language models (LLMs) have emerged as powerful tools in artificial intelligence, demonstrating remarkable capabilities in natural language processing and generation. In this article, we explore the potential applications of LLMs in enhancing cardiovascular care and research. We discuss how LLMs can be utilized to simplify complex medical information, improve patient-physician communication, and automate tasks such as summarizing medical articles and extracting key information. Additionally, we highlight the role of LLMs in categorizing and analyzing unstructured data, such as medical notes and test results, which could revolutionize data handling and interpretation in cardiovascular research. However, we also emphasize the limitations and challenges associated with LLMs, including potential biases, reasoning opacity, and the need for rigorous validation in medical contexts. This article provides a practical guide for cardiovascular professionals to understand and harness the power of LLMs while navigating their limitations. We conclude by discussing the future directions and implications of LLMs in transforming cardiovascular care and research.

Exploring the Role of Machine Learning in Diagnosing and Treating Speech Disorders: A Systematic Literature Review

Purpose

Speech disorders profoundly impact the overall quality of life by impeding social operations and hindering effective communication. This study addresses the gap in systematic reviews concerning machine learning-based assistive technology for individuals with speech disorders. The overarching purpose is to offer a comprehensive overview of the field through a Systematic Literature Review (SLR) and provide valuable insights into the landscape of ML-based solutions and related studies.

Methods

The research employs a systematic approach, utilizing a Systematic Literature Review (SLR) methodology. The study extensively examines the existing literature on machine learning-based assistive technology for speech disorders. Specific attention is given to ML techniques, characteristics of exploited datasets in the training phase, speaker languages, feature extraction techniques, and the features employed by ML algorithms.

Originality

This study contributes to the existing literature by systematically exploring the machine learning landscape in assistive technology for speech disorders. The originality lies in the focused investigation of ML-speech recognition for impaired speech disorder users over ten years (2014-2023). The emphasis on systematic research questions related to ML techniques, dataset characteristics, languages, feature extraction techniques, and feature sets adds a unique and comprehensive perspective to the current discourse.

Findings

The systematic literature review identifies significant trends and critical studies published between 2014 and 2023. In the analysis of the 65 papers from prestigious journals, support vector machines and neural networks (CNN, DNN) were the most utilized ML technique (20%, 16.92%), with the most studied disease being Dysarthria (35/65, 54% studies). Furthermore, an upsurge in using neural network-based architectures, mainly CNN and DNN, was observed after 2018. Almost half of the included studies were published between 2021 and 2022).

The clinical and imaging data fusion model for single-period cerebral CTA collateral circulation assessment

Background

The Chinese population ranks among the highest globally in terms of stroke prevalence. In the clinical diagnostic process, radiologists utilize computed tomography angiography (CTA) images for diagnosis, enabling a precise assessment of collateral circulation in the brains of stroke patients. Recent studies frequently combine imaging and machine learning methods to develop computer-aided diagnostic algorithms. However, in studies concerning collateral circulation assessment, the extracted imaging features are primarily composed of manually designed statistical features, which exhibit significant limitations in their representational capacity. Accurately assessing collateral circulation using image features in brain CTA images still presents challenges.

Methods

To tackle this issue, considering the scarcity of publicly accessible medical datasets, we combined clinical data with imaging data to establish a dataset named RadiomicsClinicCTA. Moreover, we devised two collateral circulation assessment models to exploit the synergistic potential of patients' clinical information and imaging data for a more accurate assessment of collateral circulation: data-level fusion and feature-level fusion. To remove redundant features from the dataset, we employed Levene's test and T-test methods for feature pre-screening. Subsequently, we performed feature dimensionality reduction using the LASSO and random forest algorithms and trained classification models with various machine learning algorithms on the data-level fusion dataset after feature engineering.

Results

Experimental results on the RadiomicsClinicCTA dataset demonstrate that the optimized data-level fusion model achieves an accuracy and AUC value exceeding 86% . Subsequently, we trained and assessed the performance of the feature-level fusion classification model. The results indicate the feature-level fusion classification model outperforms the optimized data-level fusion model. Comparative experiments show that the fused dataset better differentiates between good and bad side branch features relative to the pure radiomics dataset.

Conclusions

Our study underscores the efficacy of integrating clinical and imaging data through fusion models, significantly enhancing the accuracy of collateral circulation assessment in stroke patients.

Deep learning for determining the difficulty of endodontic treatment: a pilot study

BACKGROUND

To develop and validate a deep learning model for automated assessment of endodontic case difficulty from periapical radiographs.

METHODS

A dataset of 1,386 periapical radiographs was compiled from two clinical sites. Two dentists and two endodontists annotated the radiographs for difficulty using the "simple assessment" criteria from the American Association of Endodontists' case difficulty assessment form in the Endocase application. A classification task labeled cases as "easy" or "hard", while regression predicted overall difficulty scores. Convolutional neural networks (i.e. VGG16, ResNet18, ResNet50, ResNext50, and Inception v2) were used, with a baseline model trained via transfer learning from ImageNet weights. Other models was pre-trained using self-supervised contrastive learning (i.e. BYOL, SimCLR, MoCo, and DINO) on 20,295 unlabeled dental radiographs to learn representation without manual labels. Both models were evaluated using 10-fold cross-validation, with performance compared to seven human examiners (three general dentists and four endodontists) on a hold-out test set.

RESULTS

The baseline VGG16 model attained 87.62% accuracy in classifying difficulty. Self-supervised pretraining did not improve performance. Regression predicted scores with ± 3.21 score error. All models outperformed human raters, with poor inter-examiner reliability.

CONCLUSION

This pilot study demonstrated the feasibility of automated endodontic difficulty assessment via deep learning models.

Laryngeal Cancer Screening During Flexible Video Laryngoscopy Using Large Computer Vision Models

OBJECTIVE

Develop an artificial intelligence assisted computer vision model to screen for laryngeal cancer during flexible laryngoscopy.

METHODS

Using laryngeal images and flexible laryngoscopy video recordings, we developed computer vision models to classify video frames for usability and cancer screening. A separate model segments any identified lesions on the frames. We used these computer vision models to construct a video stream annotation system. This system classifies findings from flexible laryngoscopy as "potentially malignant" or "probably benign" and segments any detected lesions. Additionally, the model provides a confidence level for each classification.

RESULTS

The overall accuracy of the flexible laryngoscopy cancer screening model was 92%. For cancer screening, it achieved a sensitivity of 97.7% and a specificity of 76.9%. The segmentation model attained an average precision at a 0.50 intersection-over-union of 0.595. The confidence level for positive screening results can assist clinicians in counseling patients regarding the findings.

CONCLUSION

Our model is highly sensitive and adequately specific for laryngeal cancer screening. Segmentation helps endoscopists identify and describe potential lesions. Further optimization is required to enable the model's deployment in clinical settings for real-time annotation during flexible laryngoscopy.

Exploring the Potential of Artificial Intelligence as a Facilitating Tool for Formulation Development in Fluidized Bed Processor: a Comprehensive Review

This in-depth study looks into how artificial intelligence (AI) could be used to make formulation development easier in fluidized bed processes (FBP). FBP is complex and involves numerous variables, making optimization challenging. Various AI techniques have addressed this challenge, including machine learning, neural networks, genetic algorithms, and fuzzy logic. By integrating AI with experimental design, process modeling, and optimization strategies, intelligent systems for FBP can be developed. The advantages of AI in this context include improved process understanding, reduced time and cost, enhanced product quality, and robust formulation optimization. However, data availability, model interpretability, and regulatory compliance challenges must be addressed. Case studies demonstrate successful applications of AI in decision-making, process outcome prediction, and scale-up. AI can improve efficiency, quality, and cost-effectiveness in significant ways. Still, it is important to think carefully about data quality, how easy it is to understand, and how to follow the rules. Future research should focus on fully harnessing the potential of AI to advance formulation development in FBP.

Analysis of 3D pathology samples using weakly supervised AI

Human tissue, which is inherently three-dimensional (3D), is traditionally examined through standard-of-care histopathology as limited two-dimensional (2D) cross-sections that can insufficiently represent the tissue due to sampling bias. To holistically characterize histomorphology, 3D imaging modalities have been developed, but clinical translation is hampered by complex manual evaluation and lack of computational platforms to distill clinical insights from large, high-resolution datasets. We present TriPath, a deep-learning platform for processing tissue volumes and efficiently predicting clinical outcomes based on 3D morphological features. Recurrence risk-stratification models were trained on prostate cancer specimens imaged with open-top light-sheet microscopy or microcomputed tomography. By comprehensively capturing 3D morphologies, 3D volume-based prognostication achieves superior performance to traditional 2D slice-based approaches, including clinical/histopathological baselines from six certified genitourinary pathologists. Incorporating greater tissue volume improves prognostic performance and mitigates risk prediction variability from sampling bias, further emphasizing the value of capturing larger extents of heterogeneous morphology.

Fully automated quantitative coronary angiography versus optical coherence tomography guidance for coronary stent implantation (FLASH): Study protocol for a randomized controlled noninferiority trial

BACKGROUND

Artificial intelligence-based quantitative coronary angiography (AI-QCA) has been developed to provide a more objective and reproducible data about the severity of coronary artery stenosis and the dimensions of the vessel for intervention in real-time, overcoming the limitations of significant inter- and intraobserver variability, and time-consuming nature of on-site QCA, without requiring extra time and effort. Compared with the subjective nature of visually estimated conventional CAG guidance, AI-QCA guidance provides a more practical and standardized angiography-based approach. Although the advantage of intravascular imaging-guided PCI is increasingly recognized, their broader adoption is limited by clinical and economic barriers in many catheterization laboratories.

METHODS

The FLASH (fully automated quantitative coronary angiography versus optical coherence tomography guidance for coronary stent implantation) trial is a randomized, investigator-initiated, multicenter, open-label, noninferiority trial comparing the AI-QCA-assisted PCI strategy with optical coherence tomography-guided PCI strategy in patients with significant coronary artery disease. All operators will utilize a novel, standardized AI-QCA software and PCI protocol in the AI-QCA-assisted group. A total of 400 patients will be randomized to either group at a 1:1 ratio. The primary endpoint is the minimal stent area (mm2), determined by the final OCT run after completion of PCI. Clinical follow-up and cost-effectiveness evaluations are planned at 1 month and 6 months for all patients enrolled in the study.

RESULTS

Enrollment of a total of 400 patients from the 13 participating centers in South Korea will be completed in February 2024. Follow-up of the last enrolled patients will be completed in August 2024, and primary results will be available by late 2024.

CONCLUSION

The FLASH is the first clinical trial to evaluate the feasibility of AI-QCA-assisted PCI, and will provide the clinical evidence on AI-QCA assistance in the field of coronary intervention.

CLINICAL TRIAL REGISTRATION

URL: https://www.

CLINICALTRIALS

gov. Unique identifier: NCT05388357.

A causal perspective on brainwave modeling for brain-computer interfaces

Objective. Machine learning (ML) models have opened up enormous opportunities in the field of brain-computer Interfaces (BCIs). Despite their great success, they usually face severe limitations when they are employed in real-life applications outside a controlled laboratory setting.Approach. Mixing causal reasoning, identifying causal relationships between variables of interest, with brainwave modeling can change one's viewpoint on some of these major challenges which can be found in various stages in the ML pipeline, ranging from data collection and data pre-processing to training methods and techniques.Main results. In this work, we employ causal reasoning and present a framework aiming to breakdown and analyze important challenges of brainwave modeling for BCIs.Significance. Furthermore, we present how general ML practices as well as brainwave-specific techniques can be utilized and solve some of these identified challenges. And finally, we discuss appropriate evaluation schemes in order to measure these techniques' performance and efficiently compare them with other methods that will be developed in the future.

Artificial intelligence to enhance prehospital stroke diagnosis and triage: a perspective

As health systems organize to deliver the highest quality stroke care to their patients, there is increasing emphasis being placed on prehospital stroke recognition, accurate diagnosis, and efficient triage to improve outcomes after stroke. Emergency medical services (EMS) personnel currently rely heavily on dispatch accuracy, stroke screening tools, bypass protocols and prehospital notification to care for patients with suspected stroke, but novel tools including mobile stroke units and telemedicine-enabled ambulances are already changing the landscape of prehospital stroke care. Herein, the authors provide our perspective on the current state of prehospital stroke diagnosis and triage including several of these emerging trends. Then, we provide commentary to highlight potential artificial intelligence (AI) applications to improve stroke detection, improve accurate and timely dispatch, enhance EMS training and performance, and develop novel stroke diagnostic tools for prehospital use.

Artificial Intelligence for Multiple Sclerosis Management Using Retinal Images: Pearl, Peaks, and Pitfalls

Multiple sclerosis (MS) is a complex autoimmune disease characterized by inflammatory processes, demyelination, neurodegeneration, and axonal damage within the central nervous system (CNS). Retinal imaging, particularly Optical coherence tomography (OCT), has emerged as a crucial tool for investigating MS-related retinal injury. The integration of artificial intelligence(AI) has shown promise in enhancing OCT analysis for MS. Researchers are actively utilizing AI algorithms to accurately detect and classify MS-related abnormalities, leading to improved efficiency in diagnosis, monitoring, and personalized treatment planning. The prognostic value of AI in predicting MS disease progression has garnered substantial attention. Machine learning (ML) and deep learning (DL) algorithms can analyze longitudinal OCT data to forecast the course of the disease, providing critical information for personalized treatment planning and improved patient outcomes. Early detection of high-risk patients allows for targeted interventions to mitigate disability progression effectively. As such, AI-driven approaches yielded remarkable abilities in classifying distinct MS subtypes based on retinal features, aiding in disease characterization and guiding tailored therapeutic strategies. Additionally, these algorithms have enhanced the accuracy and efficiency of OCT image segmentation, streamlined diagnostic processes, and reduced human error. This study reviews the current research studies on the integration of AI,including ML and DL algorithms, with OCT in the context of MS. It examines the advancements, challenges, potential prospects, and ethical concerns of AI-powered techniques in enhancing MS diagnosis, monitoring disease progression, revolutionizing patient care, the development of patient screening tools, and supported clinical decision-making based on OCT images.

Convolutional Neural Network-Based Prediction of Axial Length Using Color Fundus Photography

Purpose

To develop convolutional neural network (CNN)-based models for predicting the axial length (AL) using color fundus photography (CFP) and explore associated clinical and structural characteristics.

Methods

This study enrolled 1105 fundus images from 467 participants with ALs ranging from 19.91 to 32.59 mm, obtained at National Taiwan University Hospital between 2020 and 2021. The AL measurements obtained from a scanning laser interferometer served as the gold standard. The accuracy of prediction was compared among CNN-based models with different inputs, including CFP, age, and/or sex. Heatmaps were interpreted by integrated gradients.

Results

Using age, sex, and CFP as input, the mean ± standard deviation absolute error (MAE) for AL prediction by the model was 0.771 ± 0.128 mm, outperforming models that used age and sex alone (1.263 ± 0.115 mm; P < 0.001) and CFP alone (0.831 ± 0.216 mm; P = 0.016) by 39.0% and 7.31%, respectively. The removal of relatively poor-quality CFPs resulted in a slight MAE reduction to 0.759 ± 0.120 mm without statistical significance (P = 0.24). The inclusion of age and CFP improved prediction accuracy by 5.59% (P = 0.043), while adding sex had no significant improvement (P = 0.41). The optic disc and temporal peripapillary area were highlighted as the focused areas on the heatmaps.

Conclusions

Deep learning-based prediction of AL using CFP was fairly accurate and enhanced by age inclusion. The optic disc and temporal peripapillary area may contain crucial structural information for AL prediction in CFP.

Translational Relevance

This study might aid AL assessments and the understanding of the morphologic characteristics of the fundus related to AL.

Interpretable detection of epiretinal membrane from optical coherence tomography with deep neural networks

This study aimed to automatically detect epiretinal membranes (ERM) in various OCT-scans of the central and paracentral macula region and classify them by size using deep-neural-networks (DNNs). To this end, 11,061 OCT-images were included and graded according to the presence of an ERM and its size (small 100-1000 µm, large > 1000 µm). The data set was divided into training, validation and test sets (75%, 10%, 15% of the data, respectively). An ensemble of DNNs was trained and saliency maps were generated using Guided-Backprob. OCT-scans were also transformed into a one-dimensional-value using t-SNE analysis. The DNNs' receiver-operating-characteristics on the test set showed a high performance for no-ERM, small-ERM and large-ERM cases (AUC: 0.99, 0.92, 0.99, respectively; 3-way accuracy: 89%), with small-ERMs being the most difficult ones to detect. t-SNE analysis sorted cases by size and, in particular, revealed increased classification uncertainty at the transitions between groups. Saliency maps reliably highlighted ERM, regardless of the presence of other OCT features (i.e. retinal-thickening, intraretinal pseudo-cysts, epiretinal-proliferation) and entities such as ERM-retinoschisis, macular-pseudohole and lamellar-macular-hole. This study showed therefore that DNNs can reliably detect and grade ERMs according to their size not only in the fovea but also in the paracentral region. This is also achieved in cases of hard-to-detect, small-ERMs. In addition, the generated saliency maps can be used to highlight small-ERMs that might otherwise be missed. The proposed model could be used for screening-programs or decision-support-systems in the future.

Performance evaluation of machine-assisted interpretation of Gram stains from positive blood cultures

Manual microscopy of Gram stains from positive blood cultures (PBCs) is crucial for diagnosing bloodstream infections but remains labor intensive, time consuming, and subjective. This study aimed to evaluate a scan and analysis system that combines fully automated digital microscopy with deep convolutional neural networks (CNNs) to assist the interpretation of Gram stains from PBCs for routine laboratory use. The CNN was trained to classify images of Gram stains based on staining and morphology into seven different classes: background/false-positive, Gram-positive cocci in clusters (GPCCL), Gram-positive cocci in pairs (GPCP), Gram-positive cocci in chains (GPCC), rod-shaped bacilli (RSB), yeasts, and polymicrobial specimens. A total of 1,555 Gram-stained slides of PBCs were scanned, pre-classified, and reviewed by medical professionals. The results of assisted Gram stain interpretation were compared to those of manual microscopy and cultural species identification by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS). The comparison of assisted Gram stain interpretation and manual microscopy yielded positive/negative percent agreement values of 95.8%/98.0% (GPCCL), 87.6%/99.3% (GPCP/GPCC), 97.4%/97.8% (RSB), 83.3%/99.3% (yeasts), and 87.0%/98.5% (negative/false positive). The assisted Gram stain interpretation, when compared to MALDI-TOF MS species identification, also yielded similar results. During the analytical performance study, assisted interpretation showed excellent reproducibility and repeatability. Any microorganism in PBCs should be detectable at the determined limit of detection of 105 CFU/mL. Although the CNN-based interpretation of Gram stains from PBCs is not yet ready for clinical implementation, it has potential for future integration and advancement.

Applicability of Spatial Technology in Cancer Research

This review explores spatial mapping technologies in cancer research, highlighting their crucial role in understanding the complexities of the tumor microenvironment (TME). The TME, which is an intricate ecosystem of diverse cell types, has a significant impact on tumor dynamics and treatment outcomes. This review closely examines cutting-edge spatial mapping technologies, categorizing them into capture-, imaging-, and antibody-based approaches. Each technology was scrutinized for its advantages and disadvantages, factoring in aspects such as spatial profiling area, multiplexing capabilities, and resolution. Additionally, we draw attention to the nuanced choices researchers face, with capture-based methods lending themselves to hypothesis generation, and imaging/antibody-based methods that fit neatly into hypothesis testing. Looking ahead, we anticipate a scenario in which multi-omics data are seamlessly integrated, artificial intelligence enhances data analysis, and spatiotemporal profiling opens up new dimensions.

Potential of artificial intelligence in injury prevention research and practice

There is increasing interest and use of artificial Intelligence algorithms and methods in biomedical research and practice, particularly as the technology has made significant advances in the past decade and become more accessible to more disciplines. This editorial briefly reviews this technology and its potential for injury prevention research and practice, proposing ways that it can be used to advance the discipline, as well as the potential pitfalls, concerns and biases that accompany it.

Artificial intelligence and skin cancer

Artificial intelligence is poised to rapidly reshape many fields, including that of skin cancer screening and diagnosis, both as a disruptive and assistive technology. Together with the collection and availability of large medical data sets, artificial intelligence will become a powerful tool that can be leveraged by physicians in their diagnoses and treatment plans for patients. This comprehensive review focuses on current progress toward AI applications for patients, primary care providers, dermatologists, and dermatopathologists, explores the diverse applications of image and molecular processing for skin cancer, and highlights AI's potential for patient self-screening and improving diagnostic accuracy for non-dermatologists. We additionally delve into the challenges and barriers to clinical implementation, paths forward for implementation and areas of active research.

Research in methodologies for modelling the oral cavity

The paper aims to explore the current state of understanding surrounding in silico oral modelling. This involves exploring methodologies, technologies and approaches pertaining to the modelling of the whole oral cavity; both internally and externally visible structures that may be relevant or appropriate to oral actions. Such a model could be referred to as a 'complete model' which includes consideration of a full set of facial features (i.e. not only mouth) as well as synergistic stimuli such as audio and facial thermal data. 3D modelling technologies capable of accurately and efficiently capturing a complete representation of the mouth for an individual have broad applications in the study of oral actions, due to their cost-effectiveness and time efficiency. This review delves into the field of clinical phonetics to classify oral actions pertaining to both speech and non-speech movements, identifying how the various vocal organs play a role in the articulatory and masticatory process. Vitaly, it provides a summation of 12 articulatory recording methods, forming a tool to be used by researchers in identifying which method of recording is appropriate for their work. After addressing the cost and resource-intensive limitations of existing methods, a new system of modelling is proposed that leverages external to internal correlation modelling techniques to create a more efficient models of the oral cavity. The vision is that the outcomes will be applicable to a broad spectrum of oral functions related to physiology, health and wellbeing, including speech, oral processing of foods as well as dental health. The applications may span from speech correction, designing foods for the aging population, whilst in the dental field we would be able to gain information about patient's oral actions that would become part of creating a personalised dental treatment plan.

Drug-drug interactions prediction based on deep learning and knowledge graph: A review

Drug-drug interactions (DDIs) can produce unpredictable pharmacological effects and lead to adverse events that have the potential to cause irreversible damage to the organism. Traditional methods to detect DDIs through biological or pharmacological analysis are time-consuming and expensive, therefore, there is an urgent need to develop computational methods to effectively predict drug-drug interactions. Currently, deep learning and knowledge graph techniques which can effectively extract features of entities have been widely utilized to develop DDI prediction methods. In this research, we aim to systematically review DDI prediction researches applying deep learning and graph knowledge. The available biomedical data and public databases related to drugs are firstly summarized in this review. Then, we discuss the existing drug-drug interactions prediction methods which have utilized deep learning and knowledge graph techniques and group them into three main classes: deep learning-based methods, knowledge graph-based methods, and methods that combine deep learning with knowledge graph. We comprehensively analyze the commonly used drug related data and various DDI prediction methods, and compare these prediction methods on benchmark datasets. Finally, we briefly discuss the challenges related to drug-drug interactions prediction, including asymmetric DDIs prediction and high-order DDI prediction.

Deep learning-based predictive classification of functional subpopulations of hematopoietic stem cells and multipotent progenitors

BACKGROUND

Hematopoietic stem cells (HSCs) and multipotent progenitors (MPPs) play a pivotal role in maintaining lifelong hematopoiesis. The distinction between stem cells and other progenitors, as well as the assessment of their functions, has long been a central focus in stem cell research. In recent years, deep learning has emerged as a powerful tool for cell image analysis and classification/prediction.

METHODS

In this study, we explored the feasibility of employing deep learning techniques to differentiate murine HSCs and MPPs based solely on their morphology, as observed through light microscopy (DIC) images.

RESULTS

After rigorous training and validation using extensive image datasets, we successfully developed a three-class classifier, referred to as the LSM model, capable of reliably distinguishing long-term HSCs, short-term HSCs, and MPPs. The LSM model extracts intrinsic morphological features unique to different cell types, irrespective of the methods used for cell identification and isolation, such as surface markers or intracellular GFP markers. Furthermore, employing the same deep learning framework, we created a two-class classifier that effectively discriminates between aged HSCs and young HSCs. This discovery is particularly significant as both cell types share identical surface markers yet serve distinct functions. This classifier holds the potential to offer a novel, rapid, and efficient means of assessing the functional states of HSCs, thus obviating the need for time-consuming transplantation experiments.

CONCLUSION

Our study represents the pioneering use of deep learning to differentiate HSCs and MPPs under steady-state conditions. This novel and robust deep learning-based platform will provide a basis for the future development of a new generation stem cell identification and separation system. It may also provide new insight into the molecular mechanisms underlying stem cell self-renewal.

Automatic segmentation of hepatocellular carcinoma on dynamic contrast-enhanced MRI based on deep learning

Objective. Precise hepatocellular carcinoma (HCC) detection is crucial for clinical management. While studies focus on computed tomography-based automatic algorithms, there is a rareness of research on automatic detection based on dynamic contrast enhanced (DCE) magnetic resonance imaging. This study is to develop an automatic detection and segmentation deep learning model for HCC using DCE.Approach: DCE images acquired from 2016 to 2021 were retrospectively collected. Then, 382 patients (301 male; 81 female) with 466 lesions pathologically confirmed were included and divided into an 80% training-validation set and a 20% independent test set. For external validation, 51 patients (42 male; 9 female) in another hospital from 2018 to 2021 were included. The U-net architecture was modified to accommodate multi-phasic DCE input. The model was trained with the training-validation set using five-fold cross-validation, and furtherly evaluated with the independent test set using comprehensive metrics for segmentation and detection performance. The proposed automatic segmentation model consisted of five main steps: phase registration, automatic liver region extraction using a pre-trained model, automatic HCC lesion segmentation using the multi-phasic deep learning model, ensemble of five-fold predictions, and post-processing using connected component analysis to enhance the performance to refine predictions and eliminate false positives.Main results. The proposed model achieved a mean dice similarity coefficient (DSC) of 0.81 ± 0.11, a sensitivity of 94.41 ± 15.50%, a precision of 94.19 ± 17.32%, and 0.14 ± 0.48 false positive lesions per patient in the independent test set. The model detected 88% (80/91) HCC lesions in the condition of DSC > 0.5, and the DSC per tumor was 0.80 ± 0.13. In the external set, the model detected 92% (58/62) lesions with 0.12 ± 0.33 false positives per patient, and the DSC per tumor was 0.75 ± 0.10.Significance.This study developed an automatic detection and segmentation deep learning model for HCC using DCE, which yielded promising post-processed results in accurately identifying and delineating HCC lesions.

N-of-1 medicine

ABSTRACT

The fields of precision and personalised medicine have led to promising advances in tailoring treatment to individual patients. Examples include genome/molecular alteration-guided drug selection, single-patient gene therapy design and synergy-based drug combination development, and these approaches can yield substantially diverse recommendations. Therefore, it is important to define each domain and delineate their commonalities and differences in an effort to develop novel clinical trial designs, streamline workflow development, rethink regulatory considerations, create value in healthcare and economics assessments, and other factors. These and other segments are essential to recognise the diversity within these domains to accelerate their respective workflows towards practice-changing healthcare. To emphasise these points, this article elaborates on the concept of digital health and digital medicine-enabled N-of-1 medicine, which individualises combination regimen and dosing using a patient's own data. We will conclude with recommendations for consideration when developing novel workflows based on emerging digital-based platforms.

Deep Learning for Discrimination of Hypertrophic Cardiomyopathy and Hypertensive Heart Disease on MRI Native T1 Maps

BACKGROUND

Native T1 and radiomics were used for hypertrophic cardiomyopathy (HCM) and hypertensive heart disease (HHD) differentiation previously. The current problem is that global native T1 remains modest discrimination performance and radiomics requires feature extraction beforehand. Deep learning (DL) is a promising technique in differential diagnosis. However, its feasibility for discriminating HCM and HHD has not been investigated.

PURPOSE

To examine the feasibility of DL in differentiating HCM and HHD based on T1 images and compare its diagnostic performance with other methods.

STUDY TYPE

Retrospective.

POPULATION

128 HCM patients (men, 75; age, 50 years ± 16) and 59 HHD patients (men, 40; age, 45 years ± 17).

FIELD STRENGTH/SEQUENCE

3.0T; Balanced steady-state free precession, phase-sensitive inversion recovery (PSIR) and multislice native T1 mapping.

ASSESSMENT

Compare HCM and HHD patients baseline data. Myocardial T1 values were extracted from native T1 images. Radiomics was implemented through feature extraction and Extra Trees Classifier. The DL network is ResNet32. Different input including myocardial ring (DL-myo), myocardial ring bounding box (DL-box) and the surrounding tissue without myocardial ring (DL-nomyo) were tested. We evaluate diagnostic performance through AUC of ROC curve.

STATISTICAL TESTS

Accuracy, sensitivity, specificity, ROC, and AUC were calculated. Independent t test, Mann-Whitney U-test and Chi-square test were adopted for HCM and HHD comparison. P < 0.05 was considered statistically significant.

RESULTS

DL-myo, DL-box, and DL-nomyo models showed an AUC (95% confidential interval) of 0.830 (0.702-0.959), 0.766 (0.617-0.915), 0.795 (0.654-0.936) in the testing set. AUC of native T1 and radiomics were 0.545 (0.352-0.738) and 0.800 (0.655-0.944) in the testing set.

DATA CONCLUSION

The DL method based on T1 mapping seems capable of discriminating HCM and HHD. Considering diagnostic performance, the DL network outperformed the native T1 method. Compared with radiomics, DL won an advantage for its high specificity and automated working mode.

LEVEL OF EVIDENCE

4 TECHNICAL EFFICACY STAGE: 2.

Overcoming language barriers in pediatric care: a multilingual, AI-driven curriculum for global healthcare education

Background

Online medical education often faces challenges related to communication and comprehension barriers, particularly when the instructional language differs from the healthcare providers' and caregivers' native languages. Our study addresses these challenges within pediatric healthcare by employing generative language models to produce a linguistically tailored, multilingual curriculum that covers the topics of team training, surgical procedures, perioperative care, patient journeys, and educational resources for healthcare providers and caregivers.

Methods

An interdisciplinary group formulated a video curriculum in English, addressing the nuanced challenges of pediatric healthcare. Subsequently, it was translated into Spanish, primarily emphasizing Latin American demographics, utilizing OpenAI's GPT-4. Videos were enriched with synthetic voice profiles of native speakers to uphold the consistency of the narrative.

Results

We created a collection of 45 multilingual video modules, each ranging from 3 to 8 min in length and covering essential topics such as teamwork, how to improve interpersonal communication, "How I Do It" surgical procedures, as well as focused topics in anesthesia, intensive care unit care, ward nursing, and transitions from hospital to home. Through AI-driven translation, this comprehensive collection ensures global accessibility and offers healthcare professionals and caregivers a linguistically inclusive resource for elevating standards of pediatric care worldwide.

Conclusion

This development of multilingual educational content marks a progressive step toward global standardization of pediatric care. By utilizing advanced language models for translation, we ensure that the curriculum is inclusive and accessible. This initiative aligns well with the World Health Organization's Digital Health Guidelines, advocating for digitally enabled healthcare education.

Data augmentation for invasive brain-computer interfaces based on stereo-electroencephalography (SEEG)

Objective.Deep learning is increasingly used for brain-computer interfaces (BCIs). However, the quantity of available data is sparse, especially for invasive BCIs. Data augmentation (DA) methods, such as generative models, can help to address this sparseness. However, all the existing studies on brain signals were based on convolutional neural networks and ignored the temporal dependence. This paper attempted to enhance generative models by capturing the temporal relationship from a time-series perspective.Approach. A conditional generative network (conditional transformer-based generative adversarial network (cTGAN)) based on the transformer model was proposed. The proposed method was tested using a stereo-electroencephalography (SEEG) dataset which was recorded from eight epileptic patients performing five different movements. Three other commonly used DA methods were also implemented: noise injection (NI), variational autoencoder (VAE), and conditional Wasserstein generative adversarial network with gradient penalty (cWGANGP). Using the proposed method, the artificial SEEG data was generated, and several metrics were used to compare the data quality, including visual inspection, cosine similarity (CS), Jensen-Shannon distance (JSD), and the effect on the performance of a deep learning-based classifier.Main results. Both the proposed cTGAN and the cWGANGP methods were able to generate realistic data, while NI and VAE outputted inferior samples when visualized as raw sequences and in a lower dimensional space. The cTGAN generated the best samples in terms of CS and JSD and outperformed cWGANGP significantly in enhancing the performance of a deep learning-based classifier (each of them yielding a significant improvement of 6% and 3.4%, respectively).Significance. This is the first time that DA methods have been applied to invasive BCIs based on SEEG. In addition, this study demonstrated the advantages of the model that preserves the temporal dependence from a time-series perspective.

Machine learning/artificial intelligence in sports medicine: state of the art and future directions

Machine learning (ML) is changing the way health care is practiced and recent applications of these novel statistical techniques have started to impact orthopaedic sports medicine. Machine learning enables the analysis of large volumes of data to establish complex relationships between "input" and "output" variables. These relationships may be more complex than could be established through traditional statistical analysis and can lead to the ability to predict the "output" with high levels of accuracy. Supervised learning is the most common ML approach for healthcare data and recent studies have developed algorithms to predict patient-specific outcome after surgical procedures such as hip arthroscopy and anterior cruciate ligament reconstruction. Deep learning is a higher-level ML approach that facilitates the processing and interpretation of complex datasets through artificial neural networks that are inspired by the way the human brain processes information. In orthopaedic sports medicine, deep learning has primarily been used for automatic image (computer vision) and text (natural language processing) interpretation. While applications in orthopaedic sports medicine have been increasing exponentially, one significant barrier to widespread adoption of ML remains clinician unfamiliarity with the associated methods and concepts. The goal of this review is to introduce these concepts, review current machine learning models in orthopaedic sport medicine, and discuss future opportunities for innovation within the specialty.

Real-time simultaneous refractive index and thickness mapping of sub-cellular biology at the diffraction limit

Mapping the cellular refractive index (RI) is a central task for research involving the composition of microorganisms and the development of models providing automated medical screenings with accuracy beyond 95%. These models require significantly enhancing the state-of-the-art RI mapping capabilities to provide large amounts of accurate RI data at high throughput. Here, we present a machine-learning-based technique that obtains a biological specimen's real-time RI and thickness maps from a single image acquired with a conventional color camera. This technology leverages a suitably engineered nanostructured membrane that stretches a biological analyte over its surface and absorbs transmitted light, generating complex reflection spectra from each sample point. The technique does not need pre-existing sample knowledge. It achieves 10-4 RI sensitivity and sub-nanometer thickness resolution on diffraction-limited spatial areas. We illustrate practical application by performing sub-cellular segmentation of HCT-116 colorectal cancer cells, obtaining complete three-dimensional reconstruction of the cellular regions with a characteristic length of 30 μm. These results can facilitate the development of real-time label-free technologies for biomedical studies on microscopic multicellular dynamics.

Intensive vision-guided network for radiology report generation

Objective.Automatic radiology report generation is booming due to its huge application potential for the healthcare industry. However, existing computer vision and natural language processing approaches to tackle this problem are limited in two aspects. First, when extracting image features, most of them neglect multi-view reasoning in vision and model single-view structure of medical images, such as space-view or channel-view. However, clinicians rely on multi-view imaging information for comprehensive judgment in daily clinical diagnosis. Second, when generating reports, they overlook context reasoning with multi-modal information and focus on pure textual optimization utilizing retrieval-based methods. We aim to address these two issues by proposing a model that better simulates clinicians perspectives and generates more accurate reports.Approach.Given the above limitation in feature extraction, we propose a globally-intensive attention (GIA) module in the medical image encoder to simulate and integrate multi-view vision perception. GIA aims to learn three types of vision perception: depth view, space view, and pixel view. On the other hand, to address the above problem in report generation, we explore how to involve multi-modal signals to generate precisely matched reports, i.e. how to integrate previously predicted words with region-aware visual content in next word prediction. Specifically, we design a visual knowledge-guided decoder (VKGD), which can adaptively consider how much the model needs to rely on visual information and previously predicted text to assist next word prediction. Hence, our final intensive vision-guided network framework includes a GIA-guided visual encoder and the VKGD.Main results.Experiments on two commonly-used datasets IU X-RAY and MIMIC-CXR demonstrate the superior ability of our method compared with other state-of-the-art approaches.Significance.Our model explores the potential of simulating clinicians perspectives and automatically generates more accurate reports, which promotes the exploration of medical automation and intelligence.

Deep Learning Detection of Aneurysm Clips for Magnetic Resonance Imaging Safety

Flagging the presence of metal devices before a head MRI scan is essential to allow appropriate safety checks. There is an unmet need for an automated system which can flag aneurysm clips prior to MRI appointments. We assess the accuracy with which a machine learning model can classify the presence or absence of an aneurysm clip on CT images. A total of 280 CT head scans were collected, 140 with aneurysm clips visible and 140 without. The data were used to retrain a pre-trained image classification neural network to classify CT localizer images. Models were developed using fivefold cross-validation and then tested on a holdout test set. A mean sensitivity of 100% and a mean accuracy of 82% were achieved. Predictions were explained using SHapley Additive exPlanations (SHAP), which highlighted that appropriate regions of interest were informing the models. Models were also trained from scratch to classify three-dimensional CT head scans. These did not exceed the sensitivity of the localizer models. This work illustrates an application of computer vision image classification to enhance current processes and improve patient safety.

CAD-RADS scoring of coronary CT angiography with Multi-Axis Vision Transformer: A clinically-inspired deep learning pipeline

BACKGROUND AND OBJECTIVE

The standard non-invasive imaging technique used to assess the severity and extent of Coronary Artery Disease (CAD) is Coronary Computed Tomography Angiography (CCTA). However, manual grading of each patient's CCTA according to the CAD-Reporting and Data System (CAD-RADS) scoring is time-consuming and operator-dependent, especially in borderline cases. This work proposes a fully automated, and visually explainable, deep learning pipeline to be used as a decision support system for the CAD screening procedure. The pipeline performs two classification tasks: firstly, identifying patients who require further clinical investigations and secondly, classifying patients into subgroups based on the degree of stenosis, according to commonly used CAD-RADS thresholds.

METHODS

The pipeline pre-processes multiplanar projections of the coronary arteries, extracted from the original CCTAs, and classifies them using a fine-tuned Multi-Axis Vision Transformer architecture. With the aim of emulating the current clinical practice, the model is trained to assign a per-patient score by stacking the bi-dimensional longitudinal cross-sections of the three main coronary arteries along channel dimension. Furthermore, it generates visually interpretable maps to assess the reliability of the predictions.

RESULTS

When run on a database of 1873 three-channel images of 253 patients collected at the Monzino Cardiology Center in Milan, the pipeline obtained an AUC of 0.87 and 0.93 for the two classification tasks, respectively.

CONCLUSION

According to our knowledge, this is the first model trained to assign CAD-RADS scores learning solely from patient scores and not requiring finer imaging annotation steps that are not part of the clinical routine.

Analyzing Surgical Technique in Diverse Open Surgical Videos With Multitask Machine Learning

Objective

To overcome limitations of open surgery artificial intelligence (AI) models by curating the largest collection of annotated videos and to leverage this AI-ready data set to develop a generalizable multitask AI model capable of real-time understanding of clinically significant surgical behaviors in prospectively collected real-world surgical videos.

Design, Setting, and Participants

The study team programmatically queried open surgery procedures on YouTube and manually annotated selected videos to create the AI-ready data set used to train a multitask AI model for 2 proof-of-concept studies, one generating surgical signatures that define the patterns of a given procedure and the other identifying kinematics of hand motion that correlate with surgeon skill level and experience. The Annotated Videos of Open Surgery (AVOS) data set includes 1997 videos from 23 open-surgical procedure types uploaded to YouTube from 50 countries over the last 15 years. Prospectively recorded surgical videos were collected from a single tertiary care academic medical center. Deidentified videos were recorded of surgeons performing open surgical procedures and analyzed for correlation with surgical training.

Exposures

The multitask AI model was trained on the AI-ready video data set and then retrospectively applied to the prospectively collected video data set.

Main Outcomes and Measures

Analysis of open surgical videos in near real-time, performance on AI-ready and prospectively collected videos, and quantification of surgeon skill.

Results

Using the AI-ready data set, the study team developed a multitask AI model capable of real-time understanding of surgical behaviors-the building blocks of procedural flow and surgeon skill-across space and time. Through principal component analysis, a single compound skill feature was identified, composed of a linear combination of kinematic hand attributes. This feature was a significant discriminator between experienced surgeons and surgical trainees across 101 prospectively collected surgical videos of 14 operators. For each unit increase in the compound feature value, the odds of the operator being an experienced surgeon were 3.6 times higher (95% CI, 1.67-7.62; P = .001).

Conclusions and Relevance

In this observational study, the AVOS-trained model was applied to analyze prospectively collected open surgical videos and identify kinematic descriptors of surgical skill related to efficiency of hand motion. The ability to provide AI-deduced insights into surgical structure and skill is valuable in optimizing surgical skill acquisition and ultimately improving surgical care.

External Validation of a Digital Pathology-based Multimodal Artificial Intelligence Architecture in the NRG/RTOG 9902 Phase 3 Trial

BACKGROUND

Accurate risk stratification is critical to guide management decisions in localized prostate cancer (PCa). Previously, we had developed and validated a multimodal artificial intelligence (MMAI) model generated from digital histopathology and clinical features. Here, we externally validate this model on men with high-risk or locally advanced PCa treated and followed as part of a phase 3 randomized control trial.

OBJECTIVE

To externally validate the MMAI model on men with high-risk or locally advanced PCa treated and followed as part of a phase 3 randomized control trial.

DESIGN, SETTING, AND PARTICIPANTS

Our validation cohort included 318 localized high-risk PCa patients from NRG/RTOG 9902 with available histopathology (337 [85%] of the 397 patients enrolled into the trial had available slides, of which 19 [5.6%] failed due to poor image quality).

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

Two previously locked prognostic MMAI models were validated for their intended endpoint: distant metastasis (DM) and PCa-specific mortality (PCSM). Individual clinical factors and the number of National Comprehensive Cancer Network (NCCN) high-risk features served as comparators. Subdistribution hazard ratio (sHR) was reported per standard deviation increase of the score with corresponding 95% confidence interval (CI) using Fine-Gray or Cox proportional hazards models.

RESULTS AND LIMITATIONS

The DM and PCSM MMAI algorithms were significantly and independently associated with the risk of DM (sHR [95% CI] = 2.33 [1.60-3.38], p < 0.001) and PCSM, respectively (sHR [95% CI] = 3.54 [2.38-5.28], p < 0.001) when compared against other prognostic clinical factors and NCCN high-risk features. The lower 75% of patients by DM MMAI had estimated 5- and 10-yr DM rates of 4% and 7%, and the highest quartile had average 5- and 10-yr DM rates of 19% and 32%, respectively (p < 0.001). Similar results were observed for the PCSM MMAI algorithm.

CONCLUSIONS

We externally validated the prognostic ability of MMAI models previously developed among men with localized high-risk disease. MMAI prognostic models further risk stratify beyond the clinical and pathological variables for DM and PCSM in a population of men already at a high risk for disease progression. This study provides evidence for consistent validation of our deep learning MMAI models to improve prognostication and enable more informed decision-making for patient care.

PATIENT SUMMARY

This paper presents a novel approach using images from pathology slides along with clinical variables to validate artificial intelligence (computer-generated) prognostic models. When implemented, clinicians can offer a more personalized and tailored prognostic discussion for men with localized prostate cancer.

Machine Learning Approaches for the Image-Based Identification of Surgical Wound Infections: Scoping Review

BACKGROUND

Surgical site infections (SSIs) occur frequently and impact patients and health care systems. Remote surveillance of surgical wounds is currently limited by the need for manual assessment by clinicians. Machine learning (ML)-based methods have recently been used to address various aspects of the postoperative wound healing process and may be used to improve the scalability and cost-effectiveness of remote surgical wound assessment.

OBJECTIVE

The objective of this review was to provide an overview of the ML methods that have been used to identify surgical wound infections from images.

METHODS

We conducted a scoping review of ML approaches for visual detection of SSIs following the JBI (Joanna Briggs Institute) methodology. Reports of participants in any postoperative context focusing on identification of surgical wound infections were included. Studies that did not address SSI identification, surgical wounds, or did not use image or video data were excluded. We searched MEDLINE, Embase, CINAHL, CENTRAL, Web of Science Core Collection, IEEE Xplore, Compendex, and arXiv for relevant studies in November 2022. The records retrieved were double screened for eligibility. A data extraction tool was used to chart the relevant data, which was described narratively and presented using tables. Employment of TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guidelines was evaluated and PROBAST (Prediction Model Risk of Bias Assessment Tool) was used to assess risk of bias (RoB).

RESULTS

In total, 10 of the 715 unique records screened met the eligibility criteria. In these studies, the clinical contexts and surgical procedures were diverse. All papers developed diagnostic models, though none performed external validation. Both traditional ML and deep learning methods were used to identify SSIs from mostly color images, and the volume of images used ranged from under 50 to thousands. Further, 10 TRIPOD items were reported in at least 4 studies, though 15 items were reported in fewer than 4 studies. PROBAST assessment led to 9 studies being identified as having an overall high RoB, with 1 study having overall unclear RoB.

CONCLUSIONS

Research on the image-based identification of surgical wound infections using ML remains novel, and there is a need for standardized reporting. Limitations related to variability in image capture, model building, and data sources should be addressed in the future.

Simulated outcomes for durotomy repair in minimally invasive spine surgery

Minimally invasive spine surgery (MISS) is increasingly performed using endoscopic and microscopic visualization, and the captured video can be used for surgical education and development of predictive artificial intelligence (AI) models. Video datasets depicting adverse event management are also valuable, as predictive models not exposed to adverse events may exhibit poor performance when these occur. Given that no dedicated spine surgery video datasets for AI model development are publicly available, we introduce Simulated Outcomes for Durotomy Repair in Minimally Invasive Spine Surgery (SOSpine). A validated MISS cadaveric dural repair simulator was used to educate neurosurgery residents, and surgical microscope video recordings were paired with outcome data. Objects including durotomy, needle, grasper, needle driver, and nerve hook were then annotated. Altogether, SOSpine contains 15,698 frames with 53,238 annotations and associated durotomy repair outcomes. For validation, an AI model was fine-tuned on SOSpine video and detected surgical instruments with a mean average precision of 0.77. In summary, SOSpine depicts spine surgeons managing a common complication, providing opportunities to develop surgical AI models.

Identification of difficult laryngoscopy using an optimized hybrid architecture

BACKGROUND

Identification of difficult laryngoscopy is a frequent demand in cervical spondylosis clinical surgery. This work aims to develop a hybrid architecture for identifying difficult laryngoscopy based on new indexes.

METHODS

Initially, two new indexes for identifying difficult laryngoscopy are proposed, and their efficacy for predicting difficult laryngoscopy is compared to that of two conventional indexes. Second, a hybrid adaptive architecture with convolutional layers, spatial extraction, and a vision transformer is proposed for predicting difficult laryngoscopy. The proposed adaptive hybrid architecture is then optimized by determining the optimal location for extracting spatial information.

RESULTS

The test accuracy of four indexes using simple model is 0.8320. The test accuracy of optimized hybrid architecture using four indexes is 0.8482.

CONCLUSION

The newly proposed two indexes, the angle between the lower margins of the second and sixth cervical spines and the vertical direction, are validated to be effective for recognizing difficult laryngoscopy. In addition, the optimized hybrid architecture employing four indexes demonstrates improved efficacy in detecting difficult laryngoscopy.

TRIAL REGISTRATION

Ethics permission for this research was obtained from the Medical Scientific Research Ethics Committee of Peking University Third Hospital (IRB00006761-2015021) on 30 March 2015. A well-informed agreement has been received from all participants. Patients were enrolled in this research at the Chinese Clinical Trial Registry ( http://www.chictr.org.cn , identifier: ChiCTR-ROC-16008598) on 6 June 2016.

AI co-pilot bronchoscope robot

The unequal distribution of medical resources and scarcity of experienced practitioners confine access to bronchoscopy primarily to well-equipped hospitals in developed regions, contributing to the unavailability of bronchoscopic services in underdeveloped areas. Here, we present an artificial intelligence (AI) co-pilot bronchoscope robot that empowers novice doctors to conduct lung examinations as safely and adeptly as experienced colleagues. The system features a user-friendly, plug-and-play catheter, devised for robot-assisted steering, facilitating access to bronchi beyond the fifth generation in average adult patients. Drawing upon historical bronchoscopic videos and expert imitation, our AI-human shared control algorithm enables novice doctors to achieve safe steering in the lung, mitigating misoperations. Both in vitro and in vivo results underscore that our system equips novice doctors with the skills to perform lung examinations as expertly as seasoned practitioners. This study offers innovative strategies to address the pressing issue of medical resource disparities through AI assistance.

Rapid detection of non-normal teeth on dental X-ray images using improved Mask R-CNN with attention mechanism

PURPOSE

Dental health has been getting increased attention. Timely detection of non-normal teeth (caries, residual root, retainer, teeth filling, etc.) is of great importance for people's health, well-being, and quality of life. This work proposes a rapid detection of non-normal teeth based on improved Mask R-CNN, aiming to achieve comprehensive screening of non-normal teeth on dental X-ray images.

METHODS

An improved Mask R-CNN based on attention mechanism was used to develop a non-normal teeth detection method trained on a high-quality annotated dataset, which can segment the whole mask of each non-normal tooth on the dental X-ray image immediately.

RESULTS

The average precision (AP) of the proposed non-normal teeth detection was 0.795 with an intersection-over-union of 0.5 and max detections (maxDets) of 32, which was higher than that of the typical Mask R-CNN method (AP = 0.750). In addition, validation experiments showed that the evaluation metrics (AP, recall, precision-recall (P-R) curve) of the proposed method were superior to those of the Mask R-CNN method. Furthermore, the experimental results indicated that proposed method exhibited a high sensitivity (95.65%) in detecting secondary caries. The proposed method took about 0.12 s to segment non-normal teeth on one dental X-ray image using the laptop (8G memory, NVIDIA RTX 3060 graphics processing unit), which was much faster than conventional manual methods.

CONCLUSION

The proposed method enhances the accuracy and efficiency of abnormal tooth diagnosis for practitioners, while also facilitating early detection and treatment of dental caries to substantially lower patient costs. Additionally, it can enable rapid and objective evaluation of student performance in dental examinations.

PVS-GEN: Systematic Approach for Universal Synthetic Data Generation Involving Parameterization, Verification, and Segmentation

Synthetic data generation addresses the challenges of obtaining extensive empirical datasets, offering benefits such as cost-effectiveness, time efficiency, and robust model development. Nonetheless, synthetic data-generation methodologies still encounter significant difficulties, including a lack of standardized metrics for modeling different data types and comparing generated results. This study introduces PVS-GEN, an automated, general-purpose process for synthetic data generation and verification. The PVS-GEN method parameterizes time-series data with minimal human intervention and verifies model construction using a specific metric derived from extracted parameters. For complex data, the process iteratively segments the empirical dataset until an extracted parameter can reproduce synthetic data that reflects the empirical characteristics, irrespective of the sensor data type. Moreover, we introduce the PoR metric to quantify the quality of the generated data by evaluating its time-series characteristics. Consequently, the proposed method can automatically generate diverse time-series data that covers a wide range of sensor types. We compared PVS-GEN with existing synthetic data-generation methodologies, and PVS-GEN demonstrated a superior performance. It generated data with a similarity of up to 37.1% across multiple data types and by 19.6% on average using the proposed metric, irrespective of the data type.

DTox: A deep neural network-based in visio lens for large scale toxicogenomics data

With the advancement of large-scale omics technologies, particularly transcriptomics data sets on drug and treatment response repositories available in public domain, toxicogenomics has emerged as a key field in safety pharmacology and chemical risk assessment. Traditional statistics-based bioinformatics analysis poses challenges in its application across multidimensional toxicogenomic data, including administration time, dosage, and gene expression levels. Motivated by the visual inspection workflow of field experts to augment their efficiency of screening significant genes to derive meaningful insights, together with the ability of deep neural architectures to learn the image signals, we developed DTox, a deep neural network-based in visio approach. Using the Percellome toxicogenomics database, instead of utilizing the numerical gene expression values of the transcripts (gene probes of the microarray) for dose-time combinations, DTox learned the image representation of 3D surface plots of distinct time and dosage data points to train the classifier on the experts' labels of gene probe significance. DTox outperformed statistical threshold-based bioinformatics and machine learning approaches based on numerical expression values. This result shows the ability of image-driven neural networks to overcome the limitations of classical numeric value-based approaches. Further, by augmenting the model with explainability modules, our study showed the potential to reveal the visual analysis process of human experts in toxicogenomics through the model weights. While the current work demonstrates the application of the DTox model in toxicogenomic studies, it can be further generalized as an in visio approach for multi-dimensional numeric data with applications in various fields in medical data sciences.