RESUMEN
PURPOSE: To establish a reliable machine learning model to predict malignancy in breast lesions identified by ultrasound (US) and optimize the negative predictive value to minimize unnecessary biopsies. METHODS: We included clinical and ultrasonographic attributes from 1526 breast lesions classified as BI-RADS 3, 4a, 4b, 4c, 5, and 6 that underwent US-guided breast biopsy in four institutions. We selected the most informative attributes to train nine machine learning models, ensemble models and models with tuned threshold to make inferences about the diagnosis of BI-RADS 4a and 4b lesions (validation dataset). We tested the performance of the final model with 403 new suspicious lesions. RESULTS: The most informative attributes were shape, margin, orientation and size of the lesions, the resistance index of the internal vessel, the age of the patient and the presence of a palpable lump. The highest mean negative predictive value (NPV) was achieved with the K-Nearest Neighbors algorithm (97.9%). Making ensembles did not improve the performance. Tuning the threshold did improve the performance of the models and we chose the algorithm XGBoost with the tuned threshold as the final one. The tested performance of the final model was: NPV 98.1%, false negative 1.9%, positive predictive value 77.1%, false positive 22.9%. Applying this final model, we would have missed 2 of the 231 malignant lesions of the test dataset (0.8%). CONCLUSION: Machine learning can help physicians predict malignancy in suspicious breast lesions identified by the US. Our final model would be able to avoid 60.4% of the biopsies in benign lesions missing less than 1% of the cancer cases.
RESUMEN
The purpose of this study was to retrospectively review the pathologic complete response (pCR) rate from patients (n=86) with stage II and III HER2-positive breast cancer treated with neoadjuvant chemotherapy at our institution from 2008 to 2013 and to determine possible predictive and prognostic factors. Immunohistochemistry for hormone receptors and Ki-67 was carried out. Clinical and pathological features were analyzed as predictive factors of response to therapy. For survival analysis, we used Kaplan-Meier curves to estimate 5-year survival rates and the log-rank test to compare the curves. The addition of trastuzumab to neoadjuvant chemotherapy significantly improved pCR rate from 4.8 to 46.8%, regardless of the number of preoperative trastuzumab cycles (P=0.0012). Stage II patients achieved a higher response rate compared to stage III (P=0.03). The disease-free and overall survivals were not significantly different between the group of patients that received trastuzumab in the neoadjuvant setting (56.3 and 70% at 5 years, respectively) and the group that initiated it post-operatively (75.8 and 88.7% at 5 years, respectively). Axillary pCR post neoadjuvant chemotherapy with trastuzumab was associated with reduced risk of recurrence (HR=0.34; P=0.03) and death (HR=0.21; P=0.02). In conclusion, we confirmed that trastuzumab improves pCR rates and verified that this improvement occurs even with less than four cycles of the drug. Hormone receptors and Ki-67 expressions were not predictive of response in this subset of patients. Axillary pCR clearly denotes prognosis after neoadjuvant target therapy and should be considered to be a marker of resistance, providing an opportunity to investigate new strategies for HER2-positive treatment.