In this study, our results showed that the diagnostic performance of DDC and Dapp was slightly higher than that of the conventional diffusion parameter ADC in distinguishing between malignant and benign SPL. DDC value can be used as the optimal diffusion parameter for discriminating between benign and malignant SPL.
DWI can non-invasively reflect the characteristics of biological tissues by measuring the diffusion properties of water molecules and quantitatively analyzing ADC values. We found that the ADC value of malignant SPL was significantly lower than that of benign SPL. One possible reason for this is that malignant tissue has a higher ratio of cell nucleus, nucleolus, and cytoplasm, with obvious cellular atypia, high cell density, small extracellular space, and irregularities in cell nucleus and cell membrane. These factors may result in more restricted diffusion of water molecules in both extracellular and intracellular spaces, whereas benign SPL has lower cell density and larger extracellular space, resulting in larger ADC values. Study had shown that the diffusion of water molecules in tumors is mainly affected by tumor cell density and the movement of water molecules in the stroma [15]. The cell density of malignant SPL is higher than that of benign SPL, with an increased number of cells per unit volume and a relatively reduced extracellular space, resulting in more restricted diffusion of water molecules and lower ADC values. Therefore, ADC values are a useful method for assessing histological characteristics. Some studies have shown that changes in ADC values occur earlier than morphological changes and can be used to evaluate and detect the effectiveness of tumor treatment [16, 17].
In the IVIM model, we found that the ADCslow value of malignant SPL was significantly lower than that of benign lesions, which is consistent with previous research results. Previous studies have shown that the ADCslow value of lung cancer is significantly lower than that of obstructive atelectasis. Yuan et al. found that malignant SPL has a lower ADCslow value compared to benign SPL [18]. The diffusion coefficient level largely depends on the ratio of intracellular and extracellular space in the tumor. Previous in vivo and in vitro studies have shown that the ADC value is closely related to the tumor cell structure [19–21]. This can also be inferred from our research results. Compared with benign lesions, lung cancer cells proliferate faster, showing lower ADC values and ADCslow values. However, the ADCfast and f values did not reflect the difference between benign and malignant SPL, which may be related to the instability of ADCfast and f parameters. Many factors may cause changes in IVIM parameters, such as tumor heterogeneity, technical instability, and fitting errors. Previous studies have shown that low b-value signals are more prone to measurement errors and highly sensitive to signal and noise changes, which is a challenge to obtaining a good bi-exponential model fit [22, 23]. In addition, although low b-value imaging is sensitive to vascular perfusion, other volumetric flow phenomena, such as tubular flow or glandular secretion, may cause signal attenuation and are difficult to distinguish from perfusion [24]. Moreover, the pseudo-diffusion coefficient ADCfast reflects the perfusion signal in the microvascular system, and f is the perfusion fraction characterizing the vascular volume fraction. Therefore, the variation of ADCfast and f values may be related to the heterogeneous vascular distribution in the ROI. However, in our study, the drawn ROIs were relatively small, only including a part of the lesion, and for some special benign SPL, such as sclerosing hemangioma, their own vascular components are relatively high, so the ADCfast and f values of benign SPL are more variable, leading to no statistical significance in the difference in ADCfast and f values between benign and malignant SPL. Further research is needed to use histogram analysis based on the whole volume to distinguish tumor invasiveness and tumor type, and to further evaluate the correlation between ADCfast and f values and tumor heterogeneity, which may yield more convincing results.
The stretching index model overcomes the limitations of the fast and slow diffusion compartmental models and their slow exchange in the bi-exponential model. DDC can be considered as a synthesis of various ADCs, indicating the weighted volume fraction of water molecules in the continuous distribution of ADCs within a voxel [13]. α is believed to reflect tissue heterogeneity, and a previous study [25] has shown that the heterogeneity index of malignant tumors is significantly different from that of benign tissues. Our current research indicates that DDC and α are significantly lower in malignant tumors than in benign lesions, which is similar to previous research results [26]. One possible explanation is that malignant tumor tissue exhibits more diffusion heterogeneity in vivo than benign tissue because it has more histological heterogeneity, such as cellular structural heterogeneity and tortuous angiogenesis. Similar to ADC, DDC is lower in malignant SPL due to high cell density and limited extracellular space, which restricts water molecule diffusion.
As an extension of diffusion tensor imaging (DTI), diffusion kurtosis imaging (DKI) can provide additional measurement parameter Kapp to characterize the complexity of the microenvironment by measuring the non-Gaussian diffusion behavior of water molecules in biological tissue. According to our results, both Dapp and Kapp can effectively distinguish between benign and malignant SPL, which is similar to some recent studies. The results of Das [27] and others show that DKI can distinguish between benign and malignant solitary pulmonary nodules, but DKI did not show a significant advantage compared to conventional DWI. However, it is worth noting that in Das's study, they only used three b-values (0, 500 and 1000 s/mm2), which may have biased the fitting of DKI, as it is generally believed that water molecule diffusion deviates from a Gaussian distribution only when the maximum b-value is greater than 1000 s/mm2.
It is generally believed that α can reflect the heterogeneity of microstructure, while Kapp can represent the complexity of microstructure. Although the exact meaning of α and Kapp in vivo is not fully understood, our current research results suggest that the negative correlation between them suggests that α and Kapp may be similar in distinguishing the pathological features of SPL. However, compared with the conventional ADC obtained from the standard DWI protocol, we did not observe a significant improvement in diagnostic accuracy with α and Kapp, which may be related to the selection of our ROI. In this study, the ROI we drew was relatively small, only selecting the solid component of the lesion, and the mean value was used to measure each parameter, which may not be conducive to heterogeneity evaluation. Using a whole-volume ROI histogram analysis to evaluate tumor heterogeneity and tumor type may yield better results. In addition, DeLong analysis showed no statistical differences between ADC, DDC, and Dapp in distinguishing SPL properties. For the current study, ADC has a high diagnostic accuracy (AUC = 0.904), and the DWI extension model did not show a significant advantage in distinguishing SPL properties. In practical clinical practice, using a single-index DWI model is sufficient to meet the needs of differential diagnosis.
There were several limitations in this study. Firstly, the patient population is relatively small and subtypes of SPL were not evaluated. Secondly, in this study, ROIs were selected in the solid part of the tumor rather than the entire SPL, which may have led to some selection biases due to the histological heterogeneity of SPL. In the future, further research is needed to investigate the relationship between different DWI models and histological characteristics.