In this retrospective study, there were some significant differences between PNMA and PTB patients with solitary pulmonary solid nodule in terms of qualitative and quantitative clinical data. Seven clinical variables were selected to build the clinical prediction model for differential diagnoses of PNMA from PTB, including smoking history, diabetes history, lesion location in the lower lobe, satellite lesions, cavity or vacuole, plain CT value and ΔCTV. A total of five radiomic features were selected to build the radiomic prediction model, including Flatness, Cluster Shade, Minimum, Median and Skewness. The clinical-radiomics combined model, consisting of above seven clinical features and five radiomics parameters, demonstrated good predictive ability in both the training and validation sets. Moreover, there were statistically significant differences among clinical model, radiomics model, and clinical-radiomics combined model. The clinical-radiomics combined model was better than both single models.
Smoking was an independent predictive factor to distinguish PNMA and PTB in this study. Previous study revealed that cigarette smoking increases not only the risk of progression to active pulmonary tuberculosis disease, but also the risk of new tuberculosis disease infection[18]. Another study showed that among ever smokers, a significant linear dose–response relation was observed between duration of smoking (by years) and QFT positivity (p < 0.001)[19]. It can be explained by the result of that lysosomal storage also compromises the migration of lung resident alveolar macrophages to mycobacteria suggests a mechanism for the observed susceptibility of smokers to new tuberculosis disease infection. While one study that specifically correlated smoking history with the mucinous phenotype found that mucinous adenocarcinoma had little relationship to smoking history[20]. So, we could get a result that smoking history was more common in PTB than PNMA. According to previous studies, diabetes patients are more susceptible to tuberculosis due to impaired immunity[21, 22]. And patients with diabetes mellitus showed no significant effect on the risk of lung cancer (RR: 1.10; 95% CI: 0.99–1.23; P = .087)[23]. Similar result was revealed in our present study that more than half of patients with PTB were complicated with diabetes, whereas, only few of patients with PNMA were complicated with diabetes.
Histologically, PNMA was reported to be composed mostly of mucin-rich tumor cells, with central fibrosis and alveolar spaces filled with mucin[24, 25]. PTB is formed by fibrous tissue containing caseous necrotic tissue. Because both are low-density on plain scan, and their morphological characteristics coincide greatly, one is easily misdiagnosed as the other. Our multivariate logistic regression analysis showed that the lesions location in the lower lobe with PNMA were more than PTB, whereas patients with PTB showed obvious upper lobe distribution preponderance (73.58%). Our findings were consistent with the report by Xiaoling Xu[26] and Jingping Zhang et al[27]. This difference may be attributed to the fact that the tumor cells of PNMA originate from goblet cells or columnar epithelial cells. There are relatively well differentiated cancer cells, and they can produce more mucus, which is affected by gravity. So, it was much more found in lower lobe. PTB usually occurs in the apical or posterior segment of the upper lobes and lingular segment on both lower lobes. Because of the relatively poor blood circulation (the number of macrophages is small) and the ventilation (bacteria is easy to survive), tuberculosis bacilli are more likely to stay and cause disease. The “satellite lesions” usually refers to small discrete shadows in the immediate vicinity of the main lesion. It is now widely accepted that it is the characteristic manifestation of tuberculoma. The pathological basis may be the spreading focus and fibroproliferative focus around the tuberculosis lesion. 26 of 53 PTB had satellite lesions, with a ratio of 1:2. In contrast, 5 of 124 PNMA had satellite lesions, with a ratio of 1:25. There is a high prevalence of satellite lesions in PTB, compared with PNMA, which is consistent with previous reports[27, 28]. More than half of the patients of PNMA were found to have cavity or vacuole, whereas only a quarter of patients with PTB presented this sign. Cavity appear in invasive mucinous adenocarcinomas due to incomplete obstruction of the bronchioles by mucus, resulting in alveolar hyperventilation. On the other hand, vacuole may be caused by internal necrosis of the tumor, and the necrosis is discharged through the bronchus. So, cavity or vacuole was much more found in PNMA.
In the diagnosis of lung cancer and tuberculosis, CT scans, especially the CT dynamic contrast-enhanced scan is an important method. There were many reports on CT for lung adenocarcinoma and pulmonary tuberculoma, but there were few reports about the CT features of PNMA and PTB[13, 27, 28]. As a result of this study, we determined that the mean CT value of PTB on plain scan was 31.00HU and the CT value of PNMA was 17.00HU, whereas ΔCTV was 25.49HU in the PNMA group and 4HU in the PTB group. The CT value of the PTB group on plain scan was significantly greater than the PNMA group, whereas, ΔCTV in the PTB group was lower than in the PNMA group, which is inconsistent with a former report[26]. This may be caused by the different pathologic basis of PNMA and PTB. PNMA is a mixture of mucin-rich tumor cells, fibrous tissues, and alveolar spaces filled with mucin proteins, generally with more mucin components, which may lead to low CT value. PTB is formed by fibrous tissue that contains caseous necrotic tissue with low density, but calcification can easily occur. Some calcifications are fine sand and scatter, which result in a higher CT value. According to the amount of solid component, fibrous tissue, and mucin, PNMA exhibited a complex pattern of enhancement. Furthermore, papillary or alveolar components within PNMA increased the difference between CT value. A tuberculoma's center is necrotic tissue without blood supply, its periphery is a capsule, and its inner layer is granulation tissue with blood supply. Therefore, according to the degree of caseous necrosis and granulation tissue present, enhancement is non-enhanced, annular or other forms.
To explore a much more effective method to differentiate PNMA from PTB, we extracted five independent radiomic features associated with PNMA and PTB, including Flatness, Cluster Shade, Minimum, Median and Skewness. These parameters belong to Form Factor Parameters, Texture Parameters, and Histogram Parameters. Flatness is a Form Factor Parameter which is independent of the gray-level intensity distribution in the ROI. Cluster Shade, as one of Texture Parameters, is the task of grouping a group of objects, so that objects in the same group (cluster shade) are more similar to each other (in a sense) than objects in other groups (cluster shade). The larger the Cluster Shade value, the asymmetric it is. Minimum belongs to Histogram Parameter which represent the minimum pixel value of an image (of the lesion). Median also belongs to Histogram Parameter which represent the median pixel value of an image (of the lesion). Another Histogram Parameter is skewness, which reflects the degree of asymmetry in the histogram distribution, and if the predictive value has been effective, the absolute values of the skewness would have been higher. All features above were the conversion of images to higher-dimensional data. They allowed high-throughput mining of quantitative imaging features from general medical images, followed by automated analysis to assist clinical decision-making. Previous studies revealed that radiomics features from CT could pre-operative differentiate of lung adenocarcinoma from lung tuberculoma in patients with pulmonary solitary solid nodule, also could distinguish adenocarcinomas from granulomas. To our knowledge, this is the first study to differentiate of PNMA from PTB based on radiomic features. In this study, the model based on these radiomic features also illustrated an effective role at preoperative differentiating of PNMA from PTB. According to our ROC analyses, the AUC values for the radiomics model were 0.840 and 0.960 in the training and test groups, respectively. Clinical-radiomics combined model (ROC-AUC:0.940–0.990) was significantly better than the clinical model and radiomics model. Furthermore, the decision curve analysis also demonstrated that the combined model performed significantly better than the clinical model and radiomic model in predicting outcomes. The decision curve analysis offers important information beyond the standard performance metrics of discrimination and calibration and could evaluate the clinical impact, indicating that they had a higher chance of success.
Our study had several limitations that must be taken into account. First, it was a retrospective study with a relatively small sample size, and only one institution was involved. Second, due to the study's inclusion of only patients who had pathologic results after surgery, selection bias cannot be ignored. Additionally, only PNMA and PTB were observed without calcification, so the results should be interpreted cautiously; Third, our study only evaluated only the relationship between PNMA and PTB, and other types of lung nodules need to be further explored, such as lung squamous cell carcinoma and other benign granulomatous lesions. To guide clinical practice, the model will be validated in a multicenter, prospective, large-scale study in the future, and further optimized.
In conclusion, in our present study, we established a model to differentiate PNMA from PTB by using preoperative clinicopathological features, radiomic features, and clinical-radiomic features, for the first time. The clinical-radiomic model established by us also showed good predictive value and had a potential value in clinical practice.