Enhancing Multi-disease Diagnosis of Chest X-rays with Advanced Deep-learning Networks in Real-world Data

Journal of Digital Imaging, Год журнала: 2023, Номер 36(4), С. 1332 - 1347

Опубликована: Март 29, 2023

The current artificial intelligence (AI) models are still insufficient in multi-disease diagnosis for real-world data, which always present a long-tail distribution. To tackle this issue, public dataset, "ChestX-ray14," involved fourteen (14) disease labels, was randomly divided into the train, validation, and test sets with ratios of 0.7, 0.1, 0.2. Two pretrained state-of-the-art networks, EfficientNet-b5 CoAtNet-0-rw, were chosen as backbones. After fully-connected layer, final layer 14 sigmoid activation units added to output each disease's diagnosis. achieve better adaptive learning, novel loss (Lours) designed, coalesced reweighting tail sample focus. For comparison, ResNet50 network weighted binary cross-entropy (LWBCE) used baseline, showed best performance previous study. overall individual areas under receiver operating curve (AUROC) label evaluated compared among different models. Group-score-weighted class mapping (Group-CAM) is applied visual interpretations. As result, CoAtNet-0-rw + Lours AUROC 0.842, significantly higher than LWBCE (AUROC: 0.811, p = 0.037). Group-CAM presented that model could pay proper attention lesions most labels (e.g., atelectasis, edema, effusion) but wrong other such pneumothorax; meanwhile, mislabeling dataset found. Overall, study an advanced AI diagnostic achieving significant improvement chest X-rays, particularly data challenging distributions.

Язык: Английский

---

Classification of pulmonary diseases from chest radiographs using deep transfer learning

PLoS ONE, Год журнала: 2025, Номер 20(3), С. e0316929 - e0316929

Опубликована: Март 17, 2025

Pulmonary diseases are the leading causes of disabilities and deaths worldwide. Early diagnosis pulmonary can reduce fatality rate. Chest radiographs commonly used to diagnose diseases. In clinical practice, diagnosing using chest is challenging due Overlapping complex anatomical Structures, variability in radiographs, their quality. The availability a medical specialist with extensive professional experience profoundly required. With use Convolutional Neural Networks field, be improved by automatically detecting classifying these This paper has explored effectiveness transfer learning improve predictive outcomes fifteen different radiographs. Our proposed deep learning-based computational model achieved promising results as compared existing state-of-the-art methods. reported an overall specificity 97.92%, sensitivity 97.30%, precision 97.94%, Area under Curve 97.61%. It been observed that our will valuable tool for practitioners decision-making efficiently various

Язык: Английский

---

Role of an Automated Deep Learning Algorithm for Reliable Screening of Abnormality in Chest Radiographs: A Prospective Multicenter Quality Improvement Study

Diagnostics, Год журнала: 2022, Номер 12(11), С. 2724 - 2724

Опубликована: Ноя. 7, 2022

In medical practice, chest X-rays are the most ubiquitous diagnostic imaging tests. However, current workload in extensive health care facilities and lack of well-trained radiologists is a significant challenge patient pathway. Therefore, an accurate, reliable, fast computer-aided diagnosis (CAD) system capable detecting abnormalities crucial improving radiological workflow. this prospective multicenter quality-improvement study, we have evaluated whether artificial intelligence (AI) can be used as X-ray screening tool real clinical settings. Methods: A team AI-based (qXR) part their daily reporting routine to report consecutive for multicentre study. This study took place large radiology network India between June 2021 March 2022. Results: total 65,604 were processed during period. The overall performance AI achieved normal abnormal was good. high negatively predicted value (NPV) 98.9% achieved. terms area under curve (AUC), NPV corresponding subabnormalities obtained blunted CP angle (0.97, 99.5%), hilar dysmorphism (0.86, 99.9%), cardiomegaly (0.96, 99.7%), reticulonodular pattern (0.91, rib fracture (0.98, scoliosis atelectasis calcification consolidation (0.95, 99.6%), emphysema fibrosis nodule 99.8%), opacity (0.92, 99.2%), pleural effusion pneumothorax (0.99, 99.9%). Additionally, turnaround time (TAT) decreased by about 40.63% from pre-qXR period post-qXR Conclusions: solution screened assisted ruling out patients with confidence, thus allowing focus more on assessing pathology treatment pathways.

Язык: Английский

---

AI Based Diagnosis of Pneumonia

Wireless Personal Communications, Год журнала: 2022, Номер 126(4), С. 3677 - 3692

Опубликована: Июнь 29, 2022

Язык: Английский

---

A Systematic Review: Classification of Lung Diseases from Chest X-Ray Images Using Deep Learning Algorithms

SN Computer Science, Год журнала: 2024, Номер 5(4)

Опубликована: Апрель 6, 2024

Язык: Английский

---