A benchmark of deep learning approaches to predict lung cancer risk using national lung screening trial cohort

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 11, 2025

Deep learning (DL) methods have demonstrated remarkable effectiveness in assisting with lung cancer risk prediction tasks using computed tomography (CT) scans. However, the lack of comprehensive comparison and validation state-of-the-art (SOTA) models practical settings limits their clinical application. This study aims to review analyze current SOTA deep for (malignant-benign classification). To evaluate our model's general performance, we selected 253 out 467 patients from a subset National Lung Screening Trial (NLST) who had CT scans without contrast, which are most commonly used, divided them into training test cohorts. The were preprocessed 2D-image 3D-volume formats according nodule annotations. We evaluated ten 3D eleven 2D models, pretrained on large-scale general-purpose datasets (Kinetics ImageNet) radiological (3DSeg-8, nnUnet RadImageNet), performance. Our results showed that 3D-based generally perform better than models. On cohort, best-performing model achieved an AUROC 0.86, while best reached 0.79. lowest AUROCs 0.70 0.62, respectively. Furthermore, pretraining image did not show expected performance advantage over datasets. Both can handle effectively, although superior competitors. findings highlight importance carefully selecting architectures prediction. Overall, these important implications development integration DL-based tools screening.

Язык: Английский

---

Intrusion detection and secure data storage in the cloud were recommend by a multiscale deep bidirectional gated recurrent neural network

Expert Systems with Applications, Год журнала: 2024, Номер 255, С. 124428 - 124428

Опубликована: Июнь 5, 2024

Язык: Английский

---

Enhanced brain tumor detection and segmentation using densely connected convolutional networks with stacking ensemble learning

Computers in Biology and Medicine, Год журнала: 2025, Номер 186, С. 109703 - 109703

Опубликована: Янв. 24, 2025

Язык: Английский

---

PKMT-Net: A pathological knowledge-inspired multi-scale transformer network for subtype prediction of lung cancer using histopathological images

Biomedical Signal Processing and Control, Год журнала: 2025, Номер 106, С. 107742 - 107742

Опубликована: Фев. 21, 2025

Язык: Английский

---

Optimized theory-guided convolutional neural network for lung cancer classification using CT images with advanced FPGA implementation

Biomedical Signal Processing and Control, Год журнала: 2025, Номер 106, С. 107719 - 107719

Опубликована: Фев. 22, 2025

Язык: Английский

---

Flat Window Filter-Based Sparse Fast Fourier Transform Architecture Design for MIMO-OF DM Communication System

IETE Journal of Research, Год журнала: 2025, Номер unknown, С. 1 - 9

Опубликована: Фев. 23, 2025

Язык: Английский

---

A Sensor Data-Driven Fault Diagnosis Method for Automotive Transmission Gearboxes Based on Improved EEMD and CNN-BiLSTM

Processes, Год журнала: 2025, Номер 13(4), С. 1200 - 1200

Опубликована: Апрель 16, 2025

With the rapid development of new energy vehicle technologies, higher demands have been placed on fault diagnosis for automotive transmission gearboxes. To address poor adaptability traditional methods under complex operating conditions, this paper proposes a sensor data-driven method based improved ensemble empirical mode decomposition (EEMD) combined with convolutional neural networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks. The incorporates dynamic noise adjustment mechanism, allowing amplitude to adapt characteristics signal. This improves stability accuracy signal decomposition, effectively reducing instability error accumulation associated fixed-amplitude white in EEMD. By combining CNN BiLSTM modules, approach achieves efficient feature extraction modeling. First, vibration signals gearbox different states are collected via sensors, an EEMD is employed decompose signals, removing background nonstationary components extract diagnostically significant intrinsic functions (IMFs). Then, utilized features from IMFs, deeply mining their spatiotemporal characteristics, while captures temporal sequence dependencies enhancing comprehensive modeling nonlinear features. combination these two enables adaptation achieving accurate classification identification multiple modes. Results indicate that proposed highly robust identifying modes, significantly exceeding performance conventional isolated network models. provides intelligent solution

Язык: Английский

---

Research on the Optimization Method of Project-Based Learning Design for Chinese Teaching Based on Interference-Tolerant Fast Convergence Zeroing Neural Network

International Journal of Computational Intelligence Systems, Год журнала: 2024, Номер 17(1)

Опубликована: Июль 4, 2024

Abstract The system known as project-based learning, which is applied to specific courses without compromising the coverage of necessary technical material, uses projects drive knowledge. plan and implementation learning in Chinese teaching a major project, embraces undergraduate creativity places an emphasis on real-world, open-ended are discussed this paper. In paper, research optimization method design for based optimized interference-tolerant fast convergence zeroing neural network (PBLD-ITFCZNN-BRO). It consists three stages, import phase, main stage evaluation stage. initial teacher separated students groups before lecture make sure that every group poses various traits, with some strong leadership skills hands-on skills. second phase PBL procedure helped transform what primarily passive environment (taking notes, listening, sitting) into more dynamic, student-centered, interactive one. Students presented data, articulated their concepts, then approaches problem-solving during step. teachers concluded by summarizing. performance proposed PBLD-ITFCZNN-BRO approach contains 15.26%, 20.42% 21.27% greater accuracy, 15.61%, 17.50% 20.24% precision rate, compared Investigation Computer Network Technology New Media Problem-Basis Learning Teaching Mode (CNT-PBLTM), Model Basis application Deep Physical Education Classroom Integrating Production (PBL-DL-PEC), Interdisciplinary learning: experiences reflections from electronic engineering at china (PBL-EEC) techniques, respectively.

Язык: Английский

---

RNE-DSNet: A Re-parameterization Neighborhood Enhancement-based Dual-Stream Network for CT image recognition

Engineering Science and Technology an International Journal, Год журнала: 2024, Номер 56, С. 101760 - 101760

Опубликована: Июль 10, 2024

Язык: Английский

---

CNN-Modified Encoders in U-Net for Nuclei Segmentation and Quantification of Fluorescent Images

IEEE Access, Год журнала: 2024, Номер 12, С. 107089 - 107097

Опубликована: Янв. 1, 2024

This research introduces an advanced approach to automate the segmentation and quantification of nuclei in fluorescent images through deep learning techniques. Overcoming inherent challenges such as variations pixel intensities, noisy boundaries, overlapping edges, our devised pipeline integrates U-Net architecture with state-of-the-art CNN models, EfficientNet. fusion maintains efficiency while harnessing superior capabilities Crucially, we exclusively utilize high-quality confocal generated in-house for model training, purposefully avoiding pitfalls associated publicly available synthetic data lower quality. Our training dataset encompasses over 3000 which are meticulously annotated manually ensure precision accuracy process. Additionally, post-processing is implemented refine results, providing morphological each segmented nucleus. Through comprehensive evaluation, achieves notable performance metrics, attaining F1-score 87% Intersection Union (IoU) value 80%. Furthermore, its robustness demonstrated across diverse datasets sourced from various origins, indicative broad applicability automating nucleus extraction images. innovative methodology holds significant promise advancing efforts multiple domains by facilitating a deeper understanding underlying biological processes automated analysis imagery.

Язык: Английский

---