New research suggests that a multimodal deep learning (DL) model, which incorporates ultrasound, breast magnetic resonance imaging (MRI) and clinical factors, may enhance the ability to predict the development of axillary lymph node (ALN) metastasis in patients with breast cancer.
For the retrospective study, recently published in Academic Radiology, researchers reviewed data from 465 women with breast cancer who had preoperative breast MRI and ultrasound, and subsequently assessed the multimodal DL model in an external validation cohort of 123 women.
External validation testing demonstrated an 80.9 percent area under the receiver operating characteristic curve (AUC) for the multimodal DL model in comparison to 70.5 for an ultrasound-based DL model and 70.9 percent for a MRI-based DL model.
While the multimodal model only offered a slightly higher sensitivity rate (63.83 percent) in contrast to the MRI-only (61.7 percent) and ultrasound-only models (59.6 percent), the researchers found that the multimodal model offered a greater than 10 percent higher specificity rate (85.53 percent) and more than a 20 percent higher positive predictive value (PPV) (79.27 percent) in comparison to the ultrasound- and MRI-only DL models.
“The proposed model based on (ultrasound) and MRI images outperformed the single-modality image models (DLUS and DLMRI) and the bimodal model (DLMRI+US) in terms of the AUC scores as it contained additional complementary information on the primary tumor characteristics,” wrote lead study author Xiaofeng Tang, M.D., who is affiliated with the Department of Ultrasound at Sun Yat-sen University Cancer Center and the Collaborative Innovation Center for Cancer Medicine in Guangdong, China, and colleagues.
The study authors suggested that the multimodal deep learning model could have a significant impact in axillary treatment planning for women with breast cancer.
“If the predicted ALNs are negative, there is no need for sentinel lymph node dissection (SLND). However, if the predicted ALNs are positive, SLND or axillary lymph node dissection (ALND) is needed for patients with breast cancer,” maintained Tang and colleagues.
Three Key Takeaways
1. Enhanced prediction accuracy. The multimodal deep learning (DL) model, incorporating ultrasound, breast MRI, and clinical factors, demonstrated an 80.9 percent area under the receiver operating characteristic curve (AUC), outperforming the ultrasound-only (70.5 percent) and MRI-only (70.9 percent) DL models in predicting axillary lymph node (ALN) metastasis in breast cancer patients.
2. Higher specificity and positive predictive value. The multimodal DL model achieved a specificity rate of 85.53 percent and a positive predictive value (PPV) of 79.27 percent, which were significantly higher than the rates for the single-modality models, indicating improved diagnostic accuracy for determining ALN metastasis.
3. Implications for treatment planning. The model's ability to accurately predict ALN status could impact axillary treatment planning, potentially reducing the need for sentinel lymph node dissection (SLND) or axillary lymph node dissection (ALND) in patients with negative predicted ALNs, and guiding appropriate surgical interventions for those with positive predictions.
The multimodal deep learning model may also provide a viable imaging alternative to facilitate ALN risk stratification in women with dense breasts.
“In particular, mammography is ineffective for Asian women, who typically have dense breasts. Give the effectiveness of the (ultrasound) and MRI techniques in such scenarios, the (multimodal deep learning) model is particularly suitable for this population,” added Tang and colleagues.
(Editor’s note: For related content, see “MRI-Based AI Model Shows Promise in Predicting Lymph Node Metastasis with Breast Cancer,” “MRI/CT Imaging for Axillary Lymph Nodes: What a New Breast Cancer Study Reveals” and “Ultrasound Nomogram May Enhance Prediction of Metastatic Axillary Lymph Nodes in Breast Cancer Patients.”)
Beyond the retrospective design and relatively small sample size, the authors noted they did not include peritumoral regions in their assessments for predicting ALN metastases. They also acknowledged variability in image quality for MRI and ultrasound due to the use of different systems in the study.
