Abstract This paper proposes a novel prediction model for detection of sub-centimeter pulmonary nodules from thoracic computed tomography (CT) images. We perform multi-level thresholding for lesion segmentation followed by penalised multinomial classifier for identifying the lesion type. We have considered 17,893 sub-centimeter pulmonary structures from an independent dataset collected retrospectively from Peerless Hospital, Kolkata to develop the classification model. An important feature of this dataset is class imbalance i.e. pulmonary vessels account for 70% of the structures while bones, scars, and nodules respectively account for 2%, 5%, and 23%. To the best of our knowledge, only a few researchers have addressed the issue of imbalanced classes for lung computed tomography (CT) images. The specificity, sensitivity and misclassification rates indicate that the method has the potential to reduce the number of undetected early stage lesions.

中文翻译：

---

计算机断层扫描图像中亚厘米肺结构的自动分类方法

摘要 本文提出了一种新的预测模型，用于从胸部计算机断层扫描 (CT) 图像中检测亚厘米肺结节。我们对病变分割执行多级阈值，然后是惩罚多项分类器以识别病变类型。我们考虑了从加尔各答 Peerless 医院回顾性收集的独立数据集中的 17,893 个亚厘米肺结构，以开发分类模型。该数据集的一个重要特征是类别不平衡，即肺血管占结构的 70%，而骨骼、疤痕和结节分别占 2%、5% 和 23%。据我们所知，只有少数研究人员解决了肺计算机断层扫描 (CT) 图像的类别不平衡问题。特异性，