Precise Classification of Calcified Nodule as Malignant and Benign Using Variogram and Statistical Parameters

Abstract

Background: Imaging using Computer tomography (CT) is responsible in the classification of pulmonary nodule. Identifying not only small malignant tumors but also benign solitary pulmonary nodule (SPN) is of utmost importance. This will save the patients from undergoing unneeded surgery. Certain criteria should be satisfied to classify the calcification in a benign SPN. Calcification of benign should enclose over 10% of the SPN and calcification should be Central, Diffuse, Popcorn type or Laminated. Every other patterns of calcification should not be viewed as a benign.
Methods: Based on the differential diagnosis, the features can be extracted (Statistical Parameters, variogram) and on application as feature vector to neural network (supervised/unsupervised), classification can be done. The imaging interpretation of pulmonary nodule is dependent on clinical history, size distribution and the gross representation of the nodule as well as feasibility of getting a tissue diagnosis.
Results: We have used Self Organizing Map (SOM) and Learning Vector Quantization (LVQ) as learning methods for neural network and LVQ has given precise classification results even with small number of samples.
Conclusion: The combination of variogram analysis and statistical features such as Mean, Variance, Skewness and Kurtosis has produced accurate results. The variogram analysis is not only used as tool for feature extraction in the field of Geology by geostatisticians but it has also proved its worthiness in Bio-Medical Engineering for the purpose of feature extraction.