Related Experiment Video
Updated: Jul 4, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.2K
Multi-phase deep learning framework with Multiscale Adaptive Swin Transformer and embedding attention for precision
Dhayalini M1, Revathi Alias Ponmozhi B2
1Department of ECE, School of Engineering and Technology, Dhanalakshmi Srinivasan University, Samayapuram, Tamilnadu, India. dhayalinimphd@gmail.com.
Scientific Reports
|December 13, 2025
Summary
This study introduces an advanced framework for lung nodule detection and classification, achieving high accuracy in distinguishing benign from malignant nodules. The novel approach enhances diagnostic precision for lung cancer, improving patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Healthcare
- Oncology Diagnostics
Background:
- Lung cancer is a leading cause of mortality, necessitating accurate lung nodule detection and classification.
- Current diagnostic methods face challenges in precision, scalability, and adaptability.
Purpose of the Study:
- To develop and validate an advanced multi-stage framework for lung nodule detection, segmentation, and classification.
- To improve the accuracy and efficiency of lung nodule diagnosis in clinical settings.
Main Methods:
- Utilized Sparse Edge-Preserving Enhancement (SEPE) for preprocessing.
- Employed an enhanced DeepLabv3+ architecture with ASPP and RBD for segmentation, integrating EfficientNetV2, DenseNet-201, ResNet-101, and InceptionV3 backbones.
- Implemented a Multiscale Adaptive Swin Transformer (MA-SwinT) with MEAM for nodule classification.
- Optimized hyperparameters using the Fossa Optimization Algorithm (FOA).
Main Results:
- Achieved high segmentation performance on LUNA16 (Dice: 98.75%) and LIDC-IDRI (Dice: 98.92%) datasets.
- Demonstrated superior classification accuracy (LUNA16: 99.15%, LIDC-IDRI: 99.40%) in distinguishing benign from malignant nodules.
- Framework showed high precision, recall, and F1 scores on both datasets.
Conclusions:
- The proposed multi-stage framework offers a reliable and accurate tool for lung nodule diagnosis.
- The advanced techniques enhance precision, scalability, and adaptability for clinical applications.
- Results indicate significant potential for improving lung cancer detection and patient outcomes.
