In the current study, we investigated the feasibility and accuracy of the radiomics-based prediction model for prediction of ALNM in patients with breast cancer based on DCE-MRI images of the primary tumor. Our study has three significant findings. First, the radiomics nomogram, based on age, margin, ratio, washout, and radiomics score, showed a favorable ability to discriminate between ALNM (+) and ALNM (-), with AUC values of 0.857 and 0.858 in the training and test cohorts, respectively. Second, a higher degree of kinetic heterogeneity was associated with ALNM. The multivariate analysis showed that higher values of kinetic heterogeneity (ratio [OR 6.26351] and washout [OR 5.30904]), as determined with the nomogram, were associated with ALNM in women with breast cancer. Third, in the stratification of ALNM risk in women with breast cancer and thus allow therapies to be tailored based on individual risk level. The omission of axillary lymph node dissection might be justified in this subgroup of women with low-risk group.
Breast cancer is a heterogeneous tumor with intra-tumoral temporal and spatial variation in cellularity, angiogenesis, and extravascular extracellular matrix [5]. It's a truism that the intra-tumor heterogeneity is associated with metastasis and poor prognosis due to inherently invasive biological behavior. However, it remains a challenge to quantify intra-tumoral heterogeneity in a noninvasive way before surgery. Tumor angiogenesis is one of the prerequisites for tumor progression and metastasis, and affects the uptake of contrast media within a tumor during DCE MRI. Therefore, it seems plausible that there are associations between ALNM and the breast cancer kinetic features extracted from DCE-MRI. In the current study, we hypothesized that intra-tumoral heterogeneity might be reflected in tumor enhancement kinetics from DCE-MRI and the concrete characteristic could be quantified with B.K. software. We presumed that higher tumor kinetic heterogeneity as observed on DCE-MRI images might reflect highly heterogeneous tumors with temporal and spatial variation in angiogenesis and various histopathologic components, further leading to the occurrence of ALNM.
Vascular endothelial growth factor (VEGF), as we all known, is a major stimulator of angiogenesis, which is frequently overexpressed in breast cancer [13]. Thus, the various histopathological components in angiogenesis may be reflected in the tumor enhancement kinetics features, which explains the higher kinetic heterogeneity observed in the present study. The new capillaries formed via angiogenesis are typically immature and more permeable than the normal vasculature, which may cause ALNM[14]. This could explain the higher kinetic heterogeneity value for the possibility of ALNM.
Most notably, the five DCE-MRI-based kinetic heterogeneity collected in this study showed independent predictive value in predicting ALNM, suggesting that ratio and washout could potentially identify the characteristics of intra-tumor heterogeneity driving aggressive tumor behavior. Patients assigned to the ALNM high-risk group showed increased heterogeneity, corroborating the hypothesis that tumor heterogeneity is related to aggressive tumor behavior. On the other hand, max peak enhancement, persistent, and plateau were not retained in the final model, which suggesting that DCE-MRI-based intra-tumor heterogeneity does not equal pathologic intra-tumor heterogeneity but can be used in prediction models as a valuable parameter.
The radiomics score was another independently risk factor for predicting ALNM. Establishing a radiomics score with LASSO has demonstrated excellent results in predicting lymph node metastasis in papillary thyroid carcinoma [15], cervical cancer [16], pancreatic carcinoma [17], rectal cancer [18], and lung cancer [19]. Radiomics characteristics are closely related to the microstructure and biological behavior of the tumors [16]. The radiomics score is based on the high-dimensional and statistical features, which were extracted from primary breast tumors. In this study, two radiomics features were used to calculate the radiomics score. These features represent the texture information of tumors, which is highly associated with tumor heterogeneity [20].
Age and spicule also showed a significant relationship with ALNM, which is consistent with previous research [7].The rapid tumor growth rate exceeds the nutritional capacity of its blood supply, which is a possible reason of tumor glitch. A glitch is an invasion of the cell edge caused by factors external to the cell, which is often associated with rapid tumor growth and metastasis of breast cancer.
More noteworthy was the risk stratification of the nomogram, which increased its clinical utility. For patients with low risk, we did not recommend to perform axillary lymph node dissection, but regular ultrasound follow-up, in order to reduce the surgical wound area and reduce the incidence of postoperative complications. For high-risk patients, we suggested that sentinel lymph node biopsy was not necessary and axillary lymph node dissection could be performed directly, thus reducing the intraoperative waiting time of patients and providing sufficient time for doctors to formulate reasonable preoperative surgical methods.
Our study represents a preliminary test of our underlying hypothesis and has several limitations including a retrospective design.
First, this study included a small number of patients from a single institution. Multi-institutional prospective studies would help to verify the results. Additional studies with longer follow-up and a larger population are needed to confirm the clinical utility of the nomogram. Second, the extraction of primary tumor features was semi-automatic segmentation, whether it was DCE-MRI or radiomics, which had errors to a certain extent. In the future, more prepared image acquisitions could be performed with fully automatic segmentation methods that have already been applied[21, 22]. Third, tumor hypoxia can be estimated by hypoxic imaging techniques. The correlation between tumor hypoxia and molecular markers is a potential and interesting research on ALNM. Finally, integrating DCE-MRI-based heterogeneity with tumor histopathology and molecular subtypes holds the promise for better ALNM prediction and improved clinical decision making for individual patients.