lung cancer classification

Nowadays, Histopathology images serve as a golden standard for lung cancer diagnosis since they can provide a comprehensive view of the disease and its effect on human tissue. This figure shows some representative images of squamous cell carcinoma and adenocarcinoma. Currently, pathologists make diagnosis decision based on cellular and inter-cellular level morphology. Most of current pathology diagnosis is still based on subjective opinions of pathologists and the varying abilities of doctors could result in large interpretation errors or bias. The proposed framework, which focuses on automated quantitative analysis of histopathology im- ages, could alleviate the subjectivity in NSCLC diagnosis and pro- vides supports to doctors in lung cancer classification.

The system consists of 3 modules:

1. Cell detection and segmentation

   • The process can be decomposed into eight main steps as shown above.

2. Feature extraction

   • Based on the segmented cell boundaries, seven cellular features are extracted for subsequent classification and prediction analysis. The seven features are red channel intensity, green channel intensity, blue channel intensity, gray intensity, red channel area, green channel area, and blue channel area.

3. Model training and evaluation

   • Linear and Gaussian kernel SVM classifier was trained on the extracted features. Precision, recall, and accuracy have been used as prediction performance metrics for classification.