Feasibility of Using the Privacy-preserving Large Language Model Vicuna for Labeling Radiology Reports

Radiology, Journal Year: 2023, Volume and Issue: 309(1)

Published: Oct. 1, 2023

Background Large language models (LLMs) such as ChatGPT, though proficient in many text-based tasks, are not suitable for use with radiology reports due to patient privacy constraints. Purpose To test the feasibility of using an alternative LLM (Vicuna-13B) that can be run locally labeling radiography reports. Materials and Methods Chest from MIMIC-CXR National Institutes Health (NIH) data sets were included this retrospective study. Reports examined 13 findings. Outputs reporting presence or absence findings generated by Vicuna a single-step multistep prompting strategy (prompts 1 2, respectively). Agreements between outputs CheXpert CheXbert labelers assessed Fleiss κ. Agreement three runs under hyperparameter setting introduced some randomness (temperature, 0.7) was also assessed. The performance subset 100 NIH annotated radiologist area receiver operating characteristic curve (AUC). Results A total 3269 set (median age, 68 years [IQR, 59-79 years]; 161 male patients) 25 596 47 32-58 1557 included. prompt 2 showed, on average, moderate substantial agreement (κ median, 0.57 0.45-0.66] 0.64 0.45-0.68] CheXbert) 0.52 0.41-0.65] 0.55 0.41-0.74] sets, respectively. performed at par AUC, 0.84 0.74-0.93]) both nine 11 Conclusion In proof-of-concept study, chest showed existing labelers. © RSNA, 2023 Supplemental material is available article. See editorial Cai issue.

Language: Английский

---

Automated Tool Support for Glaucoma Identification With Explainability Using Fundus Images

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 17290 - 17307

Published: Jan. 1, 2024

Glaucoma is a progressive eye condition that causes irreversible vision loss due to damage the optic nerve. Recent developments in deep learning and accessibility of computing resources have provided tool support for automated glaucoma diagnosis. Despite learning's advances disease diagnosis using medical images, generic convolutional neural networks are still not widely used practices limited trustworthiness these models. Although learning-based classification has gained popularity recent years, only few them addressed explainability interpretability models, which increases confidence such applications. This study presents state-of-the-art techniques segment classify fundus images predict conditions applies visualization explain results ease understandability. Our predictions based on U-Net with attention mechanisms ResNet50 segmentation process modified Inception V3 architecture classification. Attention backbone obtained 99.58% 98.05% accuracies disc cup segmentation, respectively RIM-ONE dataset. Additionally, we generate heatmaps highlight regions impacted both Gradient-weighted Class Activation Mapping (Grad-CAM) Grad-CAM++. model classifies segmented achieves accuracy, sensitivity, specificity values 98.97%, 99.42%, 95.59%, respectively, can be as identification images.

Language: Английский

---

Computer-Aided Diagnosis of COVID-19 from Chest X-ray Images Using Hybrid-Features and Random Forest Classifier

Healthcare, Journal Year: 2023, Volume and Issue: 11(6), P. 837 - 837

Published: March 13, 2023

In recent years, a lot of attention has been paid to using radiology imaging automatically find COVID-19. (1) Background: There are now number computer-aided diagnostic schemes that help radiologists and doctors perform COVID-19 tests quickly, accurately, consistently. (2) Methods: Using chest X-ray images, this study proposed cutting-edge scheme for the automatic recognition pneumonia. First, pre-processing method based on Gaussian filter logarithmic operator is applied input (CXR) images improve poor-quality by enhancing contrast, reducing noise, smoothing image. Second, robust features extracted from each enhanced image Convolutional Neural Network (CNNs) transformer an optimal collection grey-level co-occurrence matrices (GLCM) contain such as correlation, entropy, energy. Finally, random forest machine learning classifier used classify into three classes, COVID-19, pneumonia, or normal. The predicted output model combined with Gradient-weighted Class Activation Mapping (Grad-CAM) visualisation diagnosis. (3) Results: Our work evaluated public datasets different train-test splits (70-30%, 80-20%, 90-10%) achieved average accuracy, F1 score, recall, precision 97%, 96%, respectively. A comparative shows our outperforms existing similar work. approach can be utilised screen COVID-19-infected patients effectively. (4) Conclusions: methods also performed. For performance evaluation, metrics sensitivity, F1-measure calculated. better than methodologies, it thus effective diagnosis disease.

Language: Английский

---

A systematic review of metaheuristic algorithms in electric power systems optimization

Applied Soft Computing, Journal Year: 2023, Volume and Issue: 150, P. 111047 - 111047

Published: Nov. 11, 2023

Language: Английский

---

Lung Sound Classification With Multi-Feature Integration Utilizing Lightweight CNN Model

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 21262 - 21276

Published: Jan. 1, 2024

Detecting respiratory diseases is of utmost importance, considering that ailments represent one the most prevalent categories globally. The initial stage lung disease detection involves auscultation conducted by specialists, relying significantly on their expertise. Therefore, automating process for can yield enhanced efficiency. Artificial intelligence (AI) has shown promise in improving accuracy sound classification extracting features from sounds are relevant to task and learning relationships between these different pulmonary diseases. This paper utilizes two publicly available recordings namely, ICBHI 2017 challenge dataset another at Mendeley Data. Foremost this paper, we provide a detailed exposition about employing Convolutional Neural Network (CNN) feature extraction Mel spectrograms, frequency cepstral coefficients (MFCCs), Chromagram. highest achieved developed 91.04% 10 classes. Extending contribution, elaborates explanation model prediction Explainable Intelligence (XAI). novel contribution study CNN classifies into classes combining audio-specific enhance process.

Language: Английский

---

Deep Learning for Pneumonia Detection in Chest X-ray Images: A Comprehensive Survey

Journal of Imaging, Journal Year: 2024, Volume and Issue: 10(8), P. 176 - 176

Published: July 23, 2024

This paper addresses the significant problem of identifying relevant background and contextual literature related to deep learning (DL) as an evolving technology in order provide a comprehensive analysis application DL specific pneumonia detection via chest X-ray (CXR) imaging, which is most common cost-effective imaging technique available worldwide for diagnosis. particular key period associated with COVID-19, 2020–2023, explain, analyze, systematically evaluate limitations approaches determine their relative levels effectiveness. The context applied both aid automated substitute existing expert radiography professionals, who often have limited availability, elaborated detail. rationale undertaken research provided, along justification resources adopted relevance. explanatory text subsequent analyses are intended sufficient detail being addressed, solutions, these, ranging from more general. Indeed, our evaluation agree generally held view that use transformers, specifically, vision transformers (ViTs), promising obtaining further effective results area using CXR images. However, ViTs require extensive address several limitations, specifically following: biased datasets, data code ease model can be explained, systematic methods accurate comparison, notion class imbalance possibility adversarial attacks, latter remains fundamental research.

Language: Английский

---

Contribution to pulmonary diseases diagnostic from X-ray images using innovative deep learning models

Heliyon, Journal Year: 2024, Volume and Issue: 10(9), P. e30308 - e30308

Published: April 26, 2024

Pulmonary disease identification and characterization are among the most intriguing research topics of recent years since they require an accurate prompt diagnosis. Although pulmonary radiography has helped in lung diagnosis, interpretation radiographic image always been a major concern for doctors radiologists to reduce diagnosis errors. Due their success classification segmentation tasks, cutting-edge artificial intelligence techniques like machine learning (ML) deep (DL) widely encouraged be applied field diagnosing disorders identifying them using medical images, particularly ones. For this end, researchers concurring build systems based on these particular In paper, we proposed three deep-learning models that were trained identify presence certain diseases thoracic radiography. The first model, named "CovCXR-Net", identifies COVID-19 (two cases: or normal). second "MDCXR3-Net", pneumonia (three COVID-19, pneumonia, normal), last "MDCXR4-Net", is destined opacity (4 These have proven superiority comparison with state-of-the-art reached accuracy 99,09 %, 97.74 90,37 % respectively benchmarks.

Language: Английский

---

The effectiveness of deep learning vs. traditional methods for lung disease diagnosis using chest X-ray images: A systematic review

Applied Soft Computing, Journal Year: 2023, Volume and Issue: 147, P. 110817 - 110817

Published: Sept. 9, 2023

Language: Английский

---

Applications of Deep Learning in Trauma Radiology: A Narrative Review

Biomedical Journal, Journal Year: 2024, Volume and Issue: 48(1), P. 100743 - 100743

Published: April 26, 2024

Diagnostic imaging is essential in modern trauma care for initial evaluation and identifying injuries requiring intervention. Deep learning (DL) has become mainstream medical image analysis shown promising efficacy classification, segmentation, lesion detection. This narrative review provides the fundamental concepts developing DL algorithms presents an overview of current progress each modality. been applied to detect free fluid on Focused Assessment with Sonography Trauma (FAST), traumatic findings chest pelvic X-rays, computed tomography (CT) scans, identify intracranial hemorrhage head CT, vertebral fractures, organs like spleen, liver, lungs abdominal CT. Future directions involve expanding dataset size diversity through federated learning, enhancing model explainability transparency build clinician trust, integrating multimodal data provide more meaningful insights into injuries. Though some commercial artificial intelligence products are Food Drug Administration-approved clinical use field, adoption remains limited, highlighting need multi-disciplinary teams engineer practical, real-world solutions. Overall, shows immense potential improve efficiency accuracy imaging, but thoughtful development validation critical ensure these technologies positively impact patient care.

Language: Английский

---

Using Computer Vision Techniques to Automatically Detect Abnormalities in Chest X-rays

Diagnostics, Journal Year: 2023, Volume and Issue: 13(18), P. 2979 - 2979

Published: Sept. 18, 2023

Our research focused on creating an advanced machine-learning algorithm that accurately detects anomalies in chest X-ray images to provide healthcare professionals with a reliable tool for diagnosing various lung conditions. To achieve this, we analysed vast collection of and utilised sophisticated visual analysis techniques; such as deep learning (DL) algorithms, object recognition, categorisation models. create our model, used large training dataset X-rays, which provided valuable information visualising categorising abnormalities. We also data augmentation methods; scaling, rotation, imitation; increase the diversity training. adopted widely You Only Look Once (YOLO) v8 algorithm, recognition paradigm has demonstrated positive outcomes computer vision applications, modified it classify into distinct categories; respiratory infections, tuberculosis (TB), nodules. It was particularly effective identifying unique crucial may, otherwise, be difficult detect using traditional diagnostic methods. findings demonstrate practitioners can reliably use machine (ML) algorithms diagnose disorders greater accuracy efficiency.

Language: Английский

---