Site Editor
William J. Gradishar, MD, FACP, FASCO
William J. Gradishar, MD, FACP, FASCO
Posted: Monday, July 31, 2023
Sensitivity to neoadjuvant chemotherapy varies among patients diagnosed with early high-risk and locally advanced breast cancer. An article published in BMC Cancer presented findings from a study that developed and validated a deep learning radiopathomics model to improve pathologic complete response rates while avoiding toxic side effects. Xiaopeng Yao, PhD, of the Southwest Medical University, Luzhou, China, and colleagues reported that this model performed favorably in predicting pathologic complete response in both patient groups tested.
“When breast cancer patients have a pathologic complete response to [neoadjuvant chemotherapy], it can help patients lower the stage and shrink the tumor to receive more conservative treatment, and its event-free survival and overall survival are significantly improved. However, because of the heterogeneity and complexity of tumors, not all patients benefit from [neoadjuvant chemotherapy]…. Radiomics could predict effectively pathological complete response in patients with breast cancer,” stated Dr. Yao and colleagues.
A total of 211 patients with breast cancer who completed neoadjuvant chemotherapy were included in this retrospective study. Patients were divided into two groups: a training set (n = 155), defined as patients who performed their first treatment before September 2021, and a validation set (n = 56), which included all remaining patients. The deep learning radiopathomics model was developed using clinicopathologic features, radiomics features, and pathomics features. Three single-scale signatures were also used to comprehensively validate the model.
According to the study investigators, this radiopathomics model demonstrated a favorable performance for the prediction of pathologic complete response in both the training set (AUC = 0.933, 95% confidence interval [CI] = 0.895–0.971) and the validation set (AUC = 0.927, 95% CI = 0.858–0.996). Additionally, the model also significantly outperformed the radiomics signature (AUC = 0.821, 0.700–0.942), the pathomics signature (AUC = 0.766, 0.629–0.903), and the deep learning pathomics signature (AUC = 0.804, 0.683–0.925)—(all P < .05)—in the validation set.
Disclosure: The study authors reported no conflicts of interest.
JNCCN Spotlights
