Texture and Radiomics inspired Data-Driven Cancerous Lung Nodules Severity Classification

In the process of developing artificial consciousness to mimic human intelligence, situational decision making, and self realization, steadfast progress has been noted in the field of machine learning. The machine learning with its advantages of accommodating small dataset in the training of models accounts for its necessity in practical life. The inherent information deeply ingrained in the sea of data is extracted via radiomics and 2D texture analysis that constitutes the feature set to model the framework of the algorithm for classification, and is named as texture and radiomics features based severity classification (TRFSC). In the process, thoracic CT scans of 1018 patients in LIDC-IDRI dataset are considered for the classification of benign and malignant lung nodules. The image and annotated tumor mask information are used to extract volumetric and 2D weighted reconstructed image features. The estimated 71 features of each subject are utilized to construct the best classification model using SVM, LDA, linear regression, KNN, Bayes, and boosted trees classifiers. The performance of the classifiers are evaluated using accuracy, specificity, sensitivity, and AUC metrics. This work achieves accuracy of 0.913, specificity of 0.92, sensitivity of 0.90, and AUC of 0.96 for the SVM classifier when compared with other different classifiers. The application of various parametric values in the process of feature extraction and discrimination of classes, provides the flexibility to choose the best possible model for classification purposes. The proposed method is compared quantitatively with other classification algorithms on the ground of performance to showcase its applicability and relevance as a classification algorithm to discriminate the two benign and ma -lignant categories.

Automated summary

Significant progress has been made in the development of artificial consciousness to mimic human intelligence through machine learning. The extraction of inherent information from data using radiomics and texture analysis allows for the classification of lung nodules. The proposed method shows promising performance in classification and has potential applications in the medical field.