Multi-subtype classification model for non-small cell lung cancer based on radiomics: SLS model

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Histological subtypes of non-small cell lung cancer (NSCLC) are critical for treatment decisions.
• Previous radiomic studies primarily focused on classifying only two NSCLC subtypes (squamous cell carcinoma and adenocarcinoma).
• There is a need for multi-subtype classification models encompassing all major NSCLC types.

Purpose of the Study:

• To establish a multi-subtype classification model for the four main NSCLC subtypes: squamous cell carcinoma (SCC), adenocarcinoma (ADC), large cell carcinoma (LCC), and not otherwise specified (NOS).
• To improve classification performance and generalization ability compared to existing radiomic methods.
• To cover the entire spectrum of NSCLC subtypes.

Main Methods:

• Extracted 1029 radiomic features from CT images of 349 patients across two datasets.
• Developed a hybrid SLS model integrating synthetic minority oversampling technique, ℓ2,1-norm minimization, and support vector machines.
• Utilized a "three-in-one" concept for comprehensive subtype classification.

Main Results:

• Analyzed 247 features after dimension reduction, identifying first-order statistics, gray level co-occurrence matrix, and gray level size zone matrix as most conducive to classification.
• The SLS model achieved an average accuracy of 0.89 on the training set and 0.86 on the test set.
• Demonstrated strong classification performance for all four NSCLC subtypes.

Conclusions:

• Radiomic methods are effective for classifying NSCLC subtypes.
• The proposed SLS model accurately classifies and diagnoses the four NSCLC subtypes using CT images.
• This model holds potential for clinical application, aiding lung cancer treatment and personalized medicine.