Automatic Lung Cancer Subtypes Classification on CT Images with Self-generated Multi-modality Hybrid Features.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer subtypes require distinct therapies, making accurate diagnosis critical.
• Computed tomography (CT) is a common non-invasive tool for early lung cancer detection.
• CT-based diagnosis has limitations due to low resolution and atypical subtype presentations.

Purpose of the Study:

• To develop an automatic classification model for accurate lung cancer subtype diagnosis on CT images.
• To enhance diagnostic accuracy by incorporating histopathological information into CT analysis.
• To provide a non-invasive alternative to traditional pathological examination for lung cancer subtyping.

Main Methods:

• Proposed a novel automatic classification model for lung cancer subtypes.
• Integrated general histopathological information into CT image analysis.
• Leveraged cross-modality associations between CT and pathological images.

Main Results:

• The proposed model demonstrated superior accuracy in classifying lung cancer subtypes compared to existing CT-based models.
• Achieved significant improvements in accuracy (ACC) and area under the receiver operating characteristic curve (AUC).
• Validated performance on both internal and external evaluation datasets.

Conclusions:

• The developed model offers an effective method for automatic lung cancer subtype classification on CT images.
• Incorporating histopathological insights improves non-invasive diagnostic capabilities.
• This approach aids clinicians in accurate diagnosis, treatment planning, and prognostic management.