Radiomics based likelihood functions for cancer diagnosis

Area of Science:

• Medical Imaging
• Computational Biology
• Oncology

Background:

• Radiomic features and neural networks show promise in tumor classification.
• Incorporating discriminative cancer features into mathematical models can significantly improve classification performance.

Purpose of the Study:

• To develop and validate two radiomics-driven likelihood models for classifying lung, colon, and head and neck cancers using Computed Tomography (CT) images.

Main Methods:

• Extracted 105 3-D radiomic features from 200 lung nodules.
• Selected features using supervised and unsupervised ranking algorithms to derive diagnostic radiomic signatures.
• Integrated signatures into two mathematical likelihood functions for tumor classification.
• Validated models on 265 public datasets of lung, colon, and head and neck cancers.

Main Results:

• Achieved high classification rates for lung, colon, and head and neck cancers.
• Demonstrated the robustness of the developed radiomics-driven likelihood models.
• Confirmed the effectiveness of diagnostic mathematical functions based on general tumor phenotype.

Conclusions:

• Radiomics-driven likelihood models show significant potential for accurate cancer diagnosis.
• Mathematical functions incorporating general tumor phenotypes can be successfully developed for cancer classification.
• The developed models offer a robust approach for multi-cancer classification using CT imaging data.