BMC Cancer. 2025 Jan 13;25(1):75. doi: 10.1186/s12885-024-13423-y.

ABSTRACT

BACKGROUND: Melanoma is a highly aggressive skin cancer, where early and accurate diagnosis is crucial to improve patient outcomes. Dermoscopy, a non-invasive imaging technique, aids in melanoma detection but can be limited by subjective interpretation. Recently, machine learning and deep learning techniques have shown promise in enhancing diagnostic precision by automating the analysis of dermoscopy images.

METHODS: This systematic review examines recent advancements in machine learning (ML) and deep learning (DL) applications for melanoma diagnosis and prognosis using dermoscopy images. We conducted a thorough search across multiple databases, ultimately reviewing 34 studies published between 2016 and 2024. The review covers a range of model architectures, including DenseNet and ResNet, and discusses datasets, methodologies, and evaluation metrics used to validate model performance.

RESULTS: Our results highlight that certain deep learning architectures, such as DenseNet and DCNN demonstrated outstanding performance, achieving over 95% accuracy on the HAM10000, ISIC and other datasets for melanoma detection from dermoscopy images. The review provides insights into the strengths, limitations, and future research directions of machine learning and deep learning methods in melanoma diagnosis and prognosis. It emphasizes the challenges related to data diversity, model interpretability, and computational resource requirements.

CONCLUSION: This review underscores the potential of machine learning and deep learning methods to transform melanoma diagnosis through improved diagnostic accuracy and efficiency. Future research should focus on creating accessible, large datasets and enhancing model interpretability to increase clinical applicability. By addressing these areas, machine learning and deep learning models could play a central role in advancing melanoma diagnosis and patient care.

PMID:39806282 | DOI:10.1186/s12885-024-13423-y

Oral Radiol. 2025 Jan 14. doi: 10.1007/s11282-025-00804-7. Online ahead of print.

ABSTRACT

OBJECTIVES: Pulp stones are ectopic calcifications located in pulp tissue. The aim of this study is to introduce a novel method for detecting pulp stones on panoramic radiography images using a deep learning-based two-stage pipeline architecture.

MATERIALS AND METHODS: The first stage involved tooth localization with the YOLOv8 model, followed by pulp stone classification using ResNeXt. 375 panoramic images were included in this study, and a comprehensive set of evaluation metrics, including precision, recall, false-negative rate, false-positive rate, accuracy, and F1 score was employed to rigorously assess the performance of the proposed architecture.

RESULTS: Despite the limited annotated training data, the proposed method achieved impressive results: an accuracy of 95.4%, precision of 97.1%, recall of 96.1%, false-negative rate of 3.9%, false-positive rate of 6.1%, and a F1 score of 96.6%, outperforming existing approaches in pulp stone detection.

CONCLUSIONS: Unlike current studies, this approach adopted a more realistic scenario by utilizing a small dataset with few annotated samples, acknowledging the time-consuming and error-prone nature of expert labeling. The proposed system is particularly beneficial for dental students and newly graduated dentists who lack sufficient clinical experience, as it aids in the automatic detection of pulpal calcifications. To the best of our knowledge, this is the first study in the literature that propose a pipeline architecture to address the PS detection tasks on panoramic images.

PMID:39806222 | DOI:10.1007/s11282-025-00804-7

Eur J Hum Genet. 2025 Jan 13. doi: 10.1038/s41431-024-01782-w. Online ahead of print.

ABSTRACT

Artificial intelligence (AI) has been growing more powerful and accessible, and will increasingly impact many areas, including virtually all aspects of medicine and biomedical research. This review focuses on previous, current, and especially emerging applications of AI in clinical genetics. Topics covered include a brief explanation of different general categories of AI, including machine learning, deep learning, and generative AI. After introductory explanations and examples, the review discusses AI in clinical genetics in three main categories: clinical diagnostics; management and therapeutics; clinical support. The review concludes with short, medium, and long-term predictions about the ways that AI may affect the field of clinical genetics. Overall, while the precise speed at which AI will continue to change clinical genetics is unclear, as are the overall ramifications for patients, families, clinicians, researchers, and others, it is likely that AI will result in dramatic evolution in clinical genetics. It will be important for all those involved in clinical genetics to prepare accordingly in order to minimize the risks and maximize benefits related to the use of AI in the field.

PMID:39806188 | DOI:10.1038/s41431-024-01782-w

Surg Endosc. 2025 Jan 13. doi: 10.1007/s00464-024-11486-3. Online ahead of print.

ABSTRACT

OBJECTIVES: This study aimed to develop an automated skills assessment tool for surgical trainees using deep learning.

BACKGROUND: Optimal surgical performance in robot-assisted surgery (RAS) is essential for ensuring good surgical outcomes. This requires effective training of new surgeons, which currently relies on supervision and skill assessment by experienced surgeons. Artificial Intelligence (AI) presents an opportunity to augment existing human-based assessments.

METHODS: We used a network architecture consisting of a convolutional neural network combined with a long short-term memory (LSTM) layer to create two networks for the extraction and analysis of spatial and temporal features from video recordings of surgical procedures, facilitating action recognition and skill assessment.

RESULTS: 21 participants (16 novices and 5 experienced) performed 16 different intra-abdominal robot-assisted surgical procedures on porcine models. The action recognition network achieved an accuracy of 96.0% in identifying surgical actions. A GradCAM filter was used to enhance the model interpretability. The skill assessment network had an accuracy of 81.3% in classifying novices and experiences. Procedure plots were created to visualize the skill assessment.

CONCLUSION: Our study demonstrated that AI can be used to automate surgical action recognition and skill assessment. The use of a porcine model enables effective data collection at different levels of surgical performance, which is normally not available in the clinical setting. Future studies need to test how well AI developed within a porcine setting can be used to detect errors and provide feedback and actionable skills assessment in the clinical setting.

PMID:39806176 | DOI:10.1007/s00464-024-11486-3

Methods Mol Biol. 2025;2887:79-89. doi: 10.1007/978-1-0716-4314-3_5.

ABSTRACT

SNARE proteins play a pivotal role in membrane fusion and various cellular processes. Accurate identification of SNARE proteins is crucial for elucidating their functions in both health and disease contexts. This chapter presents a novel approach employing multiscan convolutional neural networks (CNNs) combined with position-specific scoring matrix (PSSM) profiles to accurately recognize SNARE proteins. By leveraging deep learning techniques, our method significantly enhances the accuracy and efficacy of SNARE protein classification. We detail the step-by-step methodology, including dataset preparation, feature extraction using PSI-BLAST, and the design of the multiscan CNN architecture. Our results demonstrate that this approach outperforms existing methods, providing a robust and reliable tool for bioinformatics research.

PMID:39806147 | DOI:10.1007/978-1-0716-4314-3_5

Med Biol Eng Comput. 2025 Jan 13. doi: 10.1007/s11517-024-03266-x. Online ahead of print.

ABSTRACT

In the context of chronic liver diseases, where variability in progression necessitates early and precise diagnosis, this study addresses the limitations of traditional histological analysis and the shortcomings of existing deep learning approaches. A novel patch-level classification model employing multi-scale feature extraction and fusion was developed to enhance the grading accuracy and interpretability of liver biopsies, analyzing 1322 cases across various staining methods. The study also introduces a slide-level aggregation framework, comparing different diagnostic models, to efficiently integrate local histological information. Results from extensive validation show that the slide-level model consistently achieved high F1 scores, notably 0.9 for inflammatory activity and steatosis, and demonstrated rapid diagnostic capabilities with less than one minute per slide on average. The patch-level model also performed well, with an F1 score of 0.64 for ballooning and 0.99 for other indicators, and proved transferable to public datasets. The conclusion drawn is that the proposed analytical framework offers a reliable basis for the diagnosis and treatment of chronic liver diseases, with the added benefit of robust interpretability, suggesting its practical utility in clinical settings.

PMID:39806118 | DOI:10.1007/s11517-024-03266-x

Adv Sci (Weinh). 2025 Jan 13:e2410981. doi: 10.1002/advs.202410981. Online ahead of print.

ABSTRACT

Protein phosphorylation plays a crucial role in regulating a wide range of biological processes, and its dysregulation is strongly linked to various diseases. While many phosphorylation sites have been identified so far, their functionality and regulatory effects are largely unknown. Here, a deep learning model MMFuncPhos, based on a multi-modal deep learning framework, is developed to predict functional phosphorylation sites. MMFuncPhos outperforms existing functional phosphorylation site prediction approaches. EFuncType is further developed based on transfer learning to predict whether phosphorylation of a residue upregulates or downregulates enzyme activity for the first time. The functional phosphorylation sites predicted by MMFuncPhos and the regulatory types predicted by EFuncType align with experimental findings from several newly reported protein phosphorylation studies. The study contributes to the understanding of the functional regulatory mechanism of phosphorylation and provides valuable tools for precision medicine, enzyme engineering, and drug discovery. For user convenience, these two prediction models are integrated into a web server which can be accessed at http://pkumdl.cn:8000/mmfuncphos.

PMID:39804866 | DOI:10.1002/advs.202410981

J Med Internet Res. 2025 Jan 13;27:e63004. doi: 10.2196/63004.

ABSTRACT

BACKGROUND: Digital biomarkers are increasingly used in clinical decision support for various health conditions. Speech features as digital biomarkers can offer insights into underlying physiological processes due to the complexity of speech production. This process involves respiration, phonation, articulation, and resonance, all of which rely on specific motor systems for the preparation and execution of speech. Deficits in any of these systems can cause changes in speech signal patterns. Increasing efforts are being made to develop speech-based clinical decision support systems.

OBJECTIVE: This systematic scoping review investigated the technological revolution and recent digital clinical speech signal analysis trends to understand the key concepts and research processes from clinical and technical perspectives.

METHODS: A systematic scoping review was undertaken in 6 databases guided by a set of research questions. Articles that focused on speech signal analysis for clinical decision-making were identified, and the included studies were analyzed quantitatively. A narrower scope of studies investigating neurological diseases were analyzed using qualitative content analysis.

RESULTS: A total of 389 articles met the initial eligibility criteria, of which 72 (18.5%) that focused on neurological diseases were included in the qualitative analysis. In the included studies, Parkinson disease, Alzheimer disease, and cognitive disorders were the most frequently investigated conditions. The literature explored the potential of speech feature analysis in diagnosis, differentiating between, assessing the severity and monitoring the treatment of neurological conditions. The common speech tasks used were sustained phonations, diadochokinetic tasks, reading tasks, activity-based tasks, picture descriptions, and prompted speech tasks. From these tasks, conventional speech features (such as fundamental frequency, jitter, and shimmer), advanced digital signal processing-based speech features (such as wavelet transformation-based features), and spectrograms in the form of audio images were analyzed. Traditional machine learning and deep learning approaches were used to build predictive models, whereas statistical analysis assessed variable relationships and reliability of speech features. Model evaluations primarily focused on analytical validations. A significant research gap was identified: the need for a structured research process to guide studies toward potential technological intervention in clinical settings. To address this, a research framework was proposed that adapts a design science research methodology to guide research studies systematically.

CONCLUSIONS: The findings highlight how data science techniques can enhance speech signal analysis to support clinical decision-making. By combining knowledge from clinical practice, speech science, and data science within a structured research framework, future research may achieve greater clinical relevance.

PMID:39804693 | DOI:10.2196/63004

Bioinformatics. 2025 Jan 13:btaf016. doi: 10.1093/bioinformatics/btaf016. Online ahead of print.

ABSTRACT

MOTIVATION: Recent computational approaches for predicting phage-host interaction have explored the use of sequence-only protein language models to produce embeddings of phage proteins without manual feature engineering. However, these embeddings do not directly capture protein structure information and structure-informed signals related to host specificity.

RESULTS: We present PHIStruct, a multilayer perceptron that takes in structure-aware embeddings of receptor-binding proteins, generated via the structure-aware protein language model SaProt, and then predicts the host from among the ESKAPEE genera. Compared against recent tools, PHIStruct exhibits the best balance of precision and recall, with the highest and most stable F1 score across a wide range of confidence thresholds and sequence similarity settings. The margin in performance is most pronounced when the sequence similarity between the training and test sets drops below 40%, wherein, at a relatively high-confidence threshold of above 50%, PHIStruct presents a 7% to 9% increase in class-averaged F1 over machine learning tools that do not directly incorporate structure information, as well as a 5% to 6% increase over BLASTp.

AVAILABILITY AND IMPLEMENTATION: The data and source code for our experiments and analyses are available at https://github.com/bioinfodlsu/PHIStruct.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PMID:39804673 | DOI:10.1093/bioinformatics/btaf016

Bioinformatics. 2025 Jan 13:btaf018. doi: 10.1093/bioinformatics/btaf018. Online ahead of print.

ABSTRACT

MOTIVATION: Predicting RNA-binding proteins (RBPs) is central to understanding post-transcriptional regulatory mechanisms. Here, we introduce EnrichRBP, an automated and interpretable computational platform specifically designed for the comprehensive analysis of RBP interactions with RNA.

RESULTS: EnrichRBP is a web service that enables researchers to develop original deep learning and machine learning architectures to explore the complex dynamics of RNA-binding proteins. The platform supports 70 deep learning algorithms, covering feature representation, selection, model training, comparison, optimization, and evaluation, all integrated within an automated pipeline. EnrichRBP is adept at providing comprehensive visualizations, enhancing model interpretability, and facilitating the discovery of functionally significant sequence regions crucial for RBP interactions. In addition, EnrichRBP supports base-level functional annotation tasks, offering explanations and graphical visualizations that confirm the reliability of the predicted RNA binding sites. Leveraging high-performance computing, EnrichRBP provides ultra-fast predictions ranging from seconds to hours, applicable to both pre-trained and custom model scenarios, thus proving its utility in real-world applications. Case studies highlight that EnrichRBP provides robust and interpretable predictions, demonstrating the power of deep learning in the functional analysis of RBP interactions. Finally, EnrichRBP aims to enhance the reproducibility of computational method analyses for RNA-binding protein sequences, as well as reduce the programming and hardware requirements for biologists, thereby offering meaningful functional insights.

AVAILABILITY AND IMPLEMENTATION: EnrichRBP is available at https://airbp.aibio-lab.com/. The source code is available at https://github.com/wangyb97/EnrichRBP, and detailed online documentation can be found at https://enrichrbp.readthedocs.io/en/latest/.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PMID:39804669 | DOI:10.1093/bioinformatics/btaf018

Insights Imaging. 2025 Jan 13;16(1):14. doi: 10.1186/s13244-024-01874-7.

ABSTRACT

OBJECTIVE: To develop an automatic segmentation model to delineate the adnexal masses and construct a machine learning model to differentiate between low malignant risk and intermediate-high malignant risk of adnexal masses based on ovarian-adnexal reporting and data system (O-RADS).

METHODS: A total of 663 ultrasound images of adnexal mass were collected and divided into two sets according to experienced radiologists: a low malignant risk set (n = 446) and an intermediate-high malignant risk set (n = 217). Deep learning segmentation models were trained and selected to automatically segment adnexal masses. Radiomics features were extracted utilizing a feature analysis system in Pyradiomics. Feature selection was conducted using the Spearman correlation analysis, Mann-Whitney U-test, and least absolute shrinkage and selection operator (LASSO) regression. A nomogram integrating radiomic and clinical features using a machine learning model was established and evaluated. The SHapley Additive exPlanations were used for model interpretability and visualization.

RESULTS: The FCN ResNet101 demonstrated the highest segmentation performance for adnexal masses (Dice similarity coefficient: 89.1%). Support vector machine achieved the best AUC (0.961, 95% CI: 0.925-0.996). The nomogram using the LightGBM algorithm reached the best AUC (0.966, 95% CI: 0.927-1.000). The diagnostic performance of the nomogram was comparable to that of experienced radiologists (p > 0.05) and outperformed that of less-experienced radiologists (p < 0.05). The model significantly improved the diagnostic accuracy of less-experienced radiologists.

CONCLUSIONS: The segmentation model serves as a valuable tool for the automated delineation of adnexal lesions. The machine learning model exhibited commendable classification capability and outperformed the diagnostic performance of less-experienced radiologists.

CRITICAL RELEVANCE STATEMENT: The ultrasound radiomics-based machine learning model holds the potential to elevate the professional ability of less-experienced radiologists and can be used to assist in the clinical screening of ovarian cancer.

KEY POINTS: We developed an image segmentation model to automatically delineate adnexal masses. We developed a model to classify adnexal masses based on O-RADS. The machine learning model has achieved commendable classification performance. The machine learning model possesses the capability to enhance the proficiency of less-experienced radiologists. We used SHapley Additive exPlanations to interpret and visualize the model.

PMID:39804536 | DOI:10.1186/s13244-024-01874-7

Int J Cardiovasc Imaging. 2025 Jan 13. doi: 10.1007/s10554-025-03327-8. Online ahead of print.

ABSTRACT

Coronary artery calcification (CAC) is a key marker of coronary artery disease (CAD) but is often underreported in cancer patients undergoing non-gated CT or PET/CT scans. Traditional CAC assessment requires gated CT scans, leading to increased radiation exposure and the need for specialized personnel. This study aims to develop an artificial intelligence (AI) method to automatically detect CAC from non-gated, freely-breathing, low-dose CT images obtained from positron emission tomography/computed tomography scans. A retrospective analysis of 677 PET/CT scans from a medical center was conducted. The dataset was divided into training (88%) and testing (12%) sets. The DLA-3D model was employed for high-resolution representation learning of cardiac CT images. Data preprocessing techniques were applied to normalize and augment the images. Performance was assessed using the area under the curve (AUC), accuracy, sensitivity, specificity and p-values. The AI model achieved an average AUC of 0.85 on the training set and 0.80 on the testing set. The model demonstrated expert-level performance with a specificity of 0.79, a sensitivity of 0.67, and an overall accuracy of 0.73 for the test group. In real-world scenarios, the model yielded a specificity of 0.8, sensitivity of 0.6, and an accuracy of 0.76. Comparison with human experts showed comparable performance. This study developed an AI method utilizing DLA-3D for automated CAC detection in non-gated PET/CT images. Findings indicate reliable CAC detection in routine PET/CT scans, potentially enhancing both cancer diagnosis and cardiovascular risk assessment. The DLA-3D model shows promise in aiding non-specialist physicians and may contribute to improved cardiovascular risk assessment in oncological imaging, encouraging additional CAC interpretation.

PMID:39804436 | DOI:10.1007/s10554-025-03327-8

Clin Oral Investig. 2025 Jan 13;29(1):59. doi: 10.1007/s00784-024-06136-w.

ABSTRACT

OBJECTIVES: To investigate the performance of a deep learning (DL) model for segmenting cone-beam computed tomography (CBCT) scans taken before and after mandibular horizontal guided bone regeneration (GBR) to evaluate hard tissue changes.

MATERIALS AND METHODS: The proposed SegResNet-based DL model was trained on 70 CBCT scans. It was tested on 10 pairs of pre- and post-operative CBCT scans of patients who underwent mandibular horizontal GBR. DL segmentations were compared to semi-automated (SA) segmentations of the same scans. Augmented hard tissue segmentation performance was evaluated by spatially aligning pre- and post-operative CBCT scans and subtracting preoperative segmentations obtained by DL and SA segmentations from the respective postoperative segmentations. The performance of DL compared to SA segmentation was evaluated based on the Dice similarity coefficient (DSC), intersection over the union (IoU), Hausdorff distance (HD95), and volume comparison.

RESULTS: The mean DSC and IoU between DL and SA segmentations were 0.96 ± 0.01 and 0.92 ± 0.02 in both pre- and post-operative CBCT scans. While HD95 values between DL and SA segmentations were 0.62 mm ± 0.16 mm and 0.77 mm ± 0.31 mm for pre- and post-operative CBCTs respectively. The DSC, IoU and HD95 averaged 0.85 ± 0.08; 0.78 ± 0.07 and 0.91 ± 0.92 mm for augmented hard tissue models respectively. Volumes mandible- and augmented hard tissue segmentations did not differ significantly between the DL and SA methods.

CONCLUSIONS: The SegResNet-based DL model accurately segmented CBCT scans acquired before and after mandibular horizontal GBR. However, the training database must be further increased to increase the model's robustness.

CLINICAL RELEVANCE: Automated DL segmentation could aid treatment planning for GBR and subsequent implant placement procedures and in evaluating hard tissue changes.

PMID:39804427 | DOI:10.1007/s00784-024-06136-w

J Gen Virol. 2025 Jan;106(1). doi: 10.1099/jgv.0.002067.

ABSTRACT

The complexity and speed of evolution in viruses with RNA genomes makes predictive identification of variants with epidemic or pandemic potential challenging. In recent years, machine learning has become an increasingly capable technology for addressing this challenge, as advances in methods and computational power have dramatically improved the performance of models and led to their widespread adoption across industries and disciplines. Nascent applications of machine learning technology to virus research have now expanded, providing new tools for handling large-scale datasets and leading to a reshaping of existing workflows for phenotype prediction, phylogenetic analysis, drug discovery and more. This review explores how machine learning has been applied to and has impacted the study of viruses, before addressing the strengths and limitations of its techniques and finally highlighting the next steps that are needed for the technology to reach its full potential in this challenging and ever-relevant research area.

PMID:39804261 | DOI:10.1099/jgv.0.002067

Eur Heart J. 2025 Jan 13:ehae914. doi: 10.1093/eurheartj/ehae914. Online ahead of print.

ABSTRACT

BACKGROUND AND AIMS: Current heart failure (HF) risk stratification strategies require comprehensive clinical evaluation. In this study, artificial intelligence (AI) applied to electrocardiogram (ECG) images was examined as a strategy to predict HF risk.

METHODS: Across multinational cohorts in the Yale New Haven Health System (YNHHS), UK Biobank (UKB), and Brazilian Longitudinal Study of Adult Health (ELSA-Brasil), individuals without baseline HF were followed for the first HF hospitalization. An AI-ECG model that defines cross-sectional left ventricular systolic dysfunction from 12-lead ECG images was used, and its association with incident HF was evaluated. Discrimination was assessed using Harrell's C-statistic. Pooled cohort equations to prevent HF (PCP-HF) were used as a comparator.

RESULTS: Among 231 285 YNHHS patients, 4472 had primary HF hospitalizations over 4.5 years (inter-quartile range 2.5-6.6). In UKB and ELSA-Brasil, among 42 141 and 13 454 people, 46 and 31 developed HF over 3.1 (2.1-4.5) and 4.2 (3.7-4.5) years. A positive AI-ECG screen portended a 4- to 24-fold higher risk of new-onset HF [age-, sex-adjusted hazard ratio: YNHHS, 3.88 (95% confidence interval 3.63-4.14); UKB, 12.85 (6.87-24.02); ELSA-Brasil, 23.50 (11.09-49.81)]. The association was consistent after accounting for comorbidities and the competing risk of death. Higher probabilities were associated with progressively higher HF risk. Model discrimination was 0.718 in YNHHS, 0.769 in UKB, and 0.810 in ELSA-Brasil. In YNHHS and ELSA-Brasil, incorporating AI-ECG with PCP-HF yielded a significant improvement in discrimination over PCP-HF alone.

CONCLUSIONS: An AI model applied to a single ECG image defined the risk of future HF, representing a digital biomarker for stratifying HF risk.

PMID:39804243 | DOI:10.1093/eurheartj/ehae914

J Cell Mol Med. 2025 Jan;29(1):e70351. doi: 10.1111/jcmm.70351.

ABSTRACT

Cancer is a complex disease driven by mutations in the genes that play critical roles in cellular processes. The identification of cancer driver genes is crucial for understanding tumorigenesis, developing targeted therapies and identifying rational drug targets. Experimental identification and validation of cancer driver genes are time-consuming and costly. Studies have demonstrated that interactions among genes are associated with similar phenotypes. Therefore, identifying cancer driver genes using molecular network-based approaches is necessary. Molecular network-based random walk-based approaches, which integrate mutation data with protein-protein interaction networks, have been widely employed in predicting cancer driver genes and demonstrated robust predictive potential. However, recent advancements in deep learning, particularly graph-based models, have provided novel opportunities for enhancing the prediction of cancer driver genes. This review aimed to comprehensively explore how machine learning methodologies, particularly network propagation, graph neural networks, autoencoders, graph embeddings, and attention mechanisms, improve the scalability and interpretability of molecular network-based cancer gene prediction.

PMID:39804102 | DOI:10.1111/jcmm.70351

JMIR Med Educ. 2025 Jan 10;11:e62669. doi: 10.2196/62669.

ABSTRACT

BACKGROUND: Artificial intelligence (AI) is set to shape the future of medical practice. The perspective and understanding of medical students are critical for guiding the development of educational curricula and training.

OBJECTIVE: This study aims to assess and compare medical AI-related attitudes among medical students in general medicine and in one of the visually oriented fields (pathology), along with illuminating their anticipated role of AI in the rapidly evolving landscape of AI-enhanced health care.

METHODS: This was a cross-sectional study that used a web-based survey composed of a closed-ended questionnaire. The survey addressed medical students at all educational levels across the 5 public medical schools, along with pathology residents in 4 residency programs in Jordan.

RESULTS: A total of 394 respondents participated (328 medical students and 66 pathology residents). The majority of respondents (272/394, 69%) were already aware of AI and deep learning in medicine, mainly relying on websites for information on AI, while only 14% (56/394) were aware of AI through medical schools. There was a statistically significant difference in awareness among respondents who consider themselves tech experts compared with those who do not (P=.03). More than half of the respondents believed that AI could be used to diagnose diseases automatically (213/394, 54.1% agreement), with medical students agreeing more than pathology residents (P=.04). However, more than one-third expressed fear about recent AI developments (167/394, 42.4% agreed). Two-thirds of respondents disagreed that their medical schools had educated them about AI and its potential use (261/394, 66.2% disagreed), while 46.2% (182/394) expressed interest in learning about AI in medicine. In terms of pathology-specific questions, 75.4% (297/394) agreed that AI could be used to identify pathologies in slide examinations automatically. There was a significant difference between medical students and pathology residents in their agreement (P=.001). Overall, medical students and pathology trainees had similar responses.

CONCLUSIONS: AI education should be introduced into medical school curricula to improve medical students' understanding and attitudes. Students agreed that they need to learn about AI's applications, potential hazards, and legal and ethical implications. This is the first study to analyze medical students' views and awareness of AI in Jordan, as well as the first to include pathology residents' perspectives. The findings are consistent with earlier research internationally. In comparison with prior research, these attitudes are similar in low-income and industrialized countries, highlighting the need for a global strategy to introduce AI instruction to medical students everywhere in this era of rapidly expanding technology.

PMID:39803949 | DOI:10.2196/62669

Anal Methods. 2025 Jan 13. doi: 10.1039/d4ay01970a. Online ahead of print.

ABSTRACT

Near-infrared (NIR) spectroscopy, with its advantages of non-destructive analysis, simple operation, and fast detection speed, has been widely applied in various fields. However, the effectiveness of current spectral analysis techniques still relies on complex preprocessing and feature selection of spectral data. While data-driven deep learning can automatically extract features from raw spectral data, it typically requires large amounts of labeled data for training, limiting its application in spectral analysis. To address this issue, we propose a self-supervised learning (SSL) framework based on convolutional neural networks (CNN) to enhance spectral analysis performance with small sample sizes. The method comprises two learning stages: pre-training and fine-tuning. In the pre-training stage, a large amount of pseudo-labeled data is used to learn intrinsic spectral features, followed by fine-tuning with a smaller set of labeled data to complete the final model training. Applied to our own collected dataset of three tea varieties, the proposed model achieved a classification accuracy of 99.12%. Additionally, experiments on three public datasets demonstrated that the SSL model significantly outperforms traditional machine learning methods, achieving accuracies of 97.83%, 98.14%, and 99.89%, respectively. Comparative experiments further confirmed the effectiveness of the pre-training stage, with the highest accuracy improvement, reaching 10.41%. These results highlight the potential of the proposed method for handling small sample spectral data, providing a viable solution for improved spectral analysis.

PMID:39803715 | DOI:10.1039/d4ay01970a

Am J Cancer Res. 2024 Dec 15;14(12):5885-5896. doi: 10.62347/TRNO3190. eCollection 2024.

ABSTRACT

The use of routine adjuvant radiotherapy (RT) after breast-conserving surgery (BCS) is controversial in elderly patients with early-stage breast cancer (EBC). This study aimed to evaluate the efficacy of adjuvant RT for elderly EBC patients using deep learning (DL) to personalize treatment plans. Five distinct DL models were developed to generate personalized treatment recommendations. Patients whose actual treatments aligned with the DL model suggestions were classified into the Consistent group, while those with divergent treatments were placed in the Inconsistent group. The efficacy of these models was assessed by comparing outcomes between the two groups. Multivariate logistic regression and Poisson regression analyses were used to visualize and quantify the influence of various features on adjuvant RT selection. In a cohort of 8,047 elderly EBC patients, treatment following the Deep Survival Regression with Mixture Effects (DSME) model's recommendations significantly improved survival, with inverse probability of treatment weighting (IPTW)-adjusted benefits, including a hazard ratio of 0.70 (95% CI, 0.58-0.86), a risk difference of 4.63% (95% CI, 1.59-7.66), and an extended mean survival time of 8.96 months (95% CI, 6.85-10.97), outperforming other models and the National Comprehensive Cancer Network (NCCN) guidelines. The DSME model identified elderly patients with larger tumors and more advanced disease stages as ideal candidates for adjuvant RT, though no benefit was seen in patients not recommended for it. This study introduces a novel DL-guided approach for selecting adjuvant RT in elderly EBC patients, enhancing treatment precision and potentially improving survival outcomes while minimizing unnecessary interventions.

PMID:39803647 | PMC:PMC11711541 | DOI:10.62347/TRNO3190

Heart Rhythm O2. 2024 Sep 19;5(12):862-872. doi: 10.1016/j.hroo.2024.09.007. eCollection 2024 Dec.

ABSTRACT

BACKGROUND: Detection of atrial tachyarrhythmias (ATA) on long-term electrocardiogram (ECG) recordings is a prerequisite to reduce ATA-related adverse events. However, the burden of editing massive ECG data is not sustainable. Deep learning (DL) algorithms provide improved performances on resting ECG databases. However, results on long-term Holter recordings are scarce.

OBJECTIVE: We aimed to build and evaluate a DL modular software using ECG features well known to cardiologists with a user interface that allows cardiologists to adjudicate the results and drive a second DL analysis.

METHODS: Using a large (n = 187 recordings, 249,419 one-minute samples), beat-to-beat annotated, two-lead Holter database, we built a DL algorithm with a modular structure mimicking expert physician ECG interpretation to classify atrial rhythms. The DL network includes 3 modules (cardiac rhythm regularity, electrical atrial waveform, and raw voltage by time data) followed by a decision network and a long-term weighting factor. The algorithm was validated on an external database.

RESULTS: F1 scores of our classifier were 99% for ATA detection, 95% for atrial fibrillation, and 90% for atrial flutter. Using the external Massachusetts Institute of Technology database, the classifier obtains an F1-score of 97% for the normal sinus rhythm class and 96% for the ATA class. Residual errors could be corrected by manual deactivation of 1 module in 7 of 15 of the recordings, with an accuracy < 90%.

CONCLUSION: A DL modular software using ECG features well known to cardiologists provided an excellent overall performance. Clinically significant residual errors were most often related to the classification of the atrial arrhythmia type (fibrillation vs flutter). The modular structure of the algorithm helped to edit and correct the artificial intelligence-based first-pass analysis and will provide a basis for explainability.

PMID:39803625 | PMC:PMC11721725 | DOI:10.1016/j.hroo.2024.09.007