Differentiating mass-like tuberculosis from lung cancer based on radiomics and CT features

The number of TB subtypes with irregular masses are increasing year by year, which can easily be confused with lung cancer. This study aimed to explore the value of CT radiomics analysis in differentiating mass-like tuberculosis (TB) from peripheral lung cancer.

A retrospective analysis of 37 cases with mass-like TB and 41 cases with peripheral lung cancer confirmed by pathology was performed. The performance of conventional CT (7 quantitative and 13 qualitative detection) was analyzed, and 828 texture features extracted by plain CT scan were subjected to dimensionality reduction using the minimal absolute contraction and logistic least absolute shrinkage and selection operator regression. The results were tested according to data distribution types, with differences between the TB and lung cancer groups analyzed by independent-samples t-test, Mann-Whitney test, Pearson chi-square test, or Fisher's exact test. Logistic regression was used to establish a texture feature model, a morphology model and a combined prediction model. The models' diagnostic efficacy was evaluated using receiver operating characteristic (ROC) curves.

The comparative analysis between the two groups revealed significant differences in 7 texture parameters (kurtosis, median, skewness, gray-level co-occurrence matrix, gray-level length matrix, gray-level area size matrix, and regional percentage), 4 quantitative parameters [plain scan CT value, arterial phase (AP) CT value, venous phase (VP) CT value, and the difference in CT value between the VP and plain scan], and 8 qualitative CT manifestations (lobular sign, long burr sign, exudation, pleura, necrosis, trachea, vessels, calcifications, and satellite lesions) (P<0.05); logistic regression analysis revealed the area under the ROC curve values of the texture feature, morphology, and combined prediction models to be 0.856, 0.950, and 0.982, respectively (P<0.05).

Combining morphological and radiomics models can effectively and noninvasively improve the efficiency of differentiating mass-like TB from peripheral lung cancer, which is conducive to selecting the appropriate therapy.