A total of 85 female patients (mean age, 57 years ± 13) with 85 suspicious ALNs were included in the analysis. The histopathology results of FNAB confirmed of 42 metastatic and 43 reactive lymph nodes. Patients’ ages ranged from 28 to 88 years (mean±standard deviation: 56±13) in the metastatic group, and 28 to 79 years (mean±standard deviation: 58±14) in the reactive group. The size of the lymph nodes ranged from 5.5 mm to 30.5 mm for the metastatic group and 6 mm to 30.4 mm for the reactive group. Cortical thickness ranged from 2.5 mm to 25 mm in the metastatic group and 1.4 mm to 10.6 mm in the reactive group.
Of the 85 patients included in the study, 82 had breast cancer as the primary malignancy. Among the remaining three patients, one patient with metastatic lymph node had chronic lymphocytic leukemia, and in two reactive ALN patients, no cancer was detected. Table 1 provides a summary of the clinical characteristics of the participants, along with the corresponding qHDMI, biomarkers, SWE metrics, and their respective p-values.
Table 1 Patients’ age, node characteristics, and summary of SWE and HDMI quantitative parameters.
|
Reactive(a)
N=42
|
Metastatic(a)
N=43
|
P-value(b)
|
Clinical
|
|
|
|
Age
|
58 ± 14
|
56 ± 13
|
0.40
|
Node size(mm)
|
14 ± 5.70
|
17 ± 6.50
|
0.01
|
Cortical thickness(mm)
|
4.10 ± 1.70
|
7.30 ± 4.10
|
<0.0001
|
qHDMI biomarkers
|
|
|
|
τmean
|
1.00 ± 0.03
|
1 ± 0.02
|
0.002
|
τmax
|
1.10 ± 0.16
|
1.20 ± 0.18
|
<0.0001
|
NV
|
5.80 ± 7
|
16 ± 15
|
<0.0001
|
NB
|
2.60 ± 3.70
|
8.00 ± 9.60
|
0.0004
|
VD
|
0.03 ± 0.03
|
0.05 ± 0.04
|
0.01
|
FD
|
1.00 ± 0.28
|
1.20 ± 0.29
|
0.0003
|
Dmax
|
630 ± 188
|
795 ± 168
|
0.0001
|
Dmean
|
425 ± 125
|
445 ± 72
|
0.2
|
MDmax
|
0.26 ± 0.33
|
0.50 ± 0.30
|
0.001
|
BAmax
|
56 ± 64
|
96 ± 59
|
0.007
|
SWE metrics
|
|
|
|
Emean
|
17 ± 10
|
44 ± 23
|
<0.0001
|
Emax
|
42 ± 28
|
116 ± 48
|
<0.0001
|
fmass
|
175 ± 81
|
241 ± 126
|
0.007
|
(a) Values are presented as mean ± standard deviation, (b) P-values in bold indicate statistical significance (values less 0.05)
The visual representation of the HDMI and SWE images and their corresponding B-mode ultrasound of a metastatic and a reactive ALN, along with the values of respective qHDMI biomarkers and SWE metrics, are displayed in Fig. 1. Fig. 1 (top row) shows the HDMI and SWE images of a metastatic lymph node from a woman in her 60s with grade II invasive lobular carcinoma. B-mode ultrasound image shows a large lymph node, 30 mm in largest dimension with a cortical thickness of 12.2 mm. Fig. 1 (bottom row) shows the HDMI and SWE images of a reactive lymph node in a woman in her 60s with grade 3 invasive ductal carcinoma with the corresponding qHDMI biomarkers and SWE metrics. The ultrasound features include a node size of 11.5 mm in the largest dimension with a cortical thickness of 4 mm. The hypervascularity and morphological vessel irregularity, along with the associated qHDMI biomarkers and SWE metrics, suggest this ALN is metastatic.
The distributions of all the 10 qHDMI and three SWE metrics for the metastatic and reactive ALNs are depicted in Fig. 3. SWE values (Emean, Emax, fmass) were notably higher in the metastatic group. Furthermore, qHDMI biomarkers, including FD, NB, NV, VD, Dmax, τmax, τmean, MDmax, and BAmax, were significantly higher in the metastatic nodes.
Fig.2 illustrates the Spearmean correlation coefficient between the significant SWE and qHDMI measures. All coefficients fall below 0.40 indicating low correlation among biomarkers of these two different methods.
Classification models
In our study, five distinct models (SWE, qHDMI, qHDMI-SWE, qHDMI-SWE-C, and Clinical) were trained and employed to classify malignant and benign ALNs, yielding the following outcomes: The SWE model demonstrated a sensitivity of 0.93, and a specificity of 0.91 with an AUC of 0.93. The qHDMI model, trained on the significant qHDMI parameters (NV, NB, VD, Dmax, τmean, τmax, BAmax, MDmax, and FD), exhibited a sensitivity of 0.87, a specificity of 1.00, and an AUC of 0.97. The sensitivity and specificity of the qHDMI-SWE model were found to be 0.93 and 1.00, respectively, with an AUC of 0.97. The qHDMI-SWE-C model integrated qHDMI, SWE, and clinical parameters. This model displayed a sensitivity of 0.87, a specificity of 1.00, and an AUC of 0.98. Finally, the Clinical model, trained only on age, maximum diameter of ALN, and ALN cortical thickness, achieved a sensitivity of 40% and a specificity of 91%, with an AUC of 0.62. Fig. 4 demonstrates all models’ ROC curves, and the table beneath the plots contains performance metrics of the models.
Table 2 shows the pair-wise AUC comparisons between the five developed models using the p-values obtained from the DeLong’s test. The results indicate that compared to the baseline Clinical model, all the models exhibit improved classification performance. On the other hand, even though the qHDMI model had higher AUC than the SWE model (0.97 compared to 0.93), and the addition of clinical features to the combination of SWE and qHDMI biomarkers led to a slight increase in the AUC (up to 0.98), these improvements were not statistically significant.
Table 2: P-values of AUC pair-wise comparisons using DeLong test.
Model (AUC)
|
Clinical (0.62)
|
SWE (0.93)
|
qHDMI (0.97)
|
qHDMI-SWE (0.97)
|
qHDMI-SWE-C (0.98)
|
Clinical (0.62)
|
1.0000
|
0.0100
|
0.0011
|
0.0017
|
0.0007
|
SWE (0.93)
|
0.0100
|
1.0000
|
0.5780
|
0.2044
|
0.3110
|
qHDMI (0.97)
|
0.0011
|
0.5780
|
1.0000
|
0.8754
|
0.6619
|
qHDMI-SWE (0.97)
|
0.0017
|
0.2044
|
0.8754
|
1.0000
|
0.7314
|
qHDMI-SWE-C (0.98)
|
0.0007
|
0.3110
|
0.6619
|
0.7314
|
1.0000
|
P-values in bold indicate statistical significance (values less 0.05)
The visual representation and quantitative measures of the qHDMI and SWE images for a metastatic and a reactive ALN with false negative and false positive SWE outcomes are displayed in Fig. 5. The qHDMI and SWE images of a metastatic lymph node from a woman in her 60s with grade II invasive ductal carcinoma are shown in Fig. 5, top row. B-mode ultrasound image of this node shows a lymph node of 5.5 mm in largest dimension with a cortical thickness of 3.6 mm. While visual presentation and quantitative biomarkers of qHDMI suggest this ALN as metastatic, the SWE map and estimates incorrectly favored this ALN as reactive. Furthermore, the HDMI images and quantitative biomarkers of a reactive lymph node with a size of 10 mm in largest dimension and with a cortical thickness of 4.1 mm, from a woman in her 50s with grade II invasive ductal carcinoma suggest a reactive ALN, but the SWE map estimating high stiffness falsely suggests a metastatic ALN (Fig. 5, bottom row).
Breast cancer subtypes
In this study, we also conducted an analysis to investigate associations between breast cancer immunohistochemical subtypes luminal A and B and qHDMI and SWE biomarkers of ALNs. Specifically, we investigated distributional differences of the qHDMI and SWE biomarkers between ALNs with these two subtypes as the primary cancer. Among our 81 breast cancer patients, we could gather immunohistochemical data from 67 patients. 21 patients were Luminal A, 41 were Luminal B, 2 were HER2+, and three were triple-negative (TNBC). Since luminal A and B comprised most of the sample, we did the analysis only for the comparison of the luminal A and B groups.
Fig. 6 offers an overview of the set of biomarkers that displayed statistically significant distributional differences for the two different luminal subtypes, identified using the Wilcoxon rank-sum test. The more aggressive luminal subtype B corresponded to higher elasticity values (Emax and Emean) and qHDMI biomarkers (NV, MDmax, τmax, and FD).