Related Experiment Video
Updated: Jun 8, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Enhanced NSCLC subtyping and staging through attention-augmented multi-task deep learning: A novel diagnostic tool.
Runhuang Yang1, Weiming Li1, Siqi Yu1
1Department of Epidemiology and Health Statistics, School of Public Health, Capital Medical University, Beijing, China; Beijing Municipal Key Laboratory of Clinical Epidemiology, Capital Medical University, Beijing, China.
International Journal of Medical Informatics
|November 8, 2024
Summary
This study introduces an attention-enhanced multi-task learning model for non-small cell lung cancer (NSCLC) classification. The novel approach significantly improves the accuracy of histologic subtype and clinical stage identification in NSCLC patients.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Non-small cell lung cancer (NSCLC) diagnosis relies on accurate histologic subtyping and clinical staging.
- Current deep learning models face challenges in accurately classifying NSCLC subtypes and stages.
Purpose of the Study:
- To develop a novel multi-task learning approach with attention encoders for NSCLC classification.
- To achieve superior performance in classifying histologic subtypes and clinical stages of NSCLC compared to existing deep-learning models.
Main Methods:
- Collected data from six public TCIA datasets, comprising 4548 CT slices from 758 NSCLC patients.
- Evaluated multi-task learning models integrating attention mechanisms for feature extraction and classification.
- Utilized convolution-based and attention-based modules in a sequential manner for task-specific branches.
Main Results:
- The MobileNet-based multi-task learning model with attention (MN-MTL-A) achieved superior performance.
- MN-MTL-A obtained Area Under the Curve (AUC) scores of 0.963 for subtype classification and 0.966 for staging.
- The model significantly outperformed models without attention and single-task learning models (P<0.01).
Conclusions:
- Integrating attention encoder blocks significantly enhanced NSCLC histologic subtyping and clinical staging accuracy.
- The proposed model shows potential for clinical application due to reduced reliance on radiologist annotation.
