Value of radiomics based on CE-MRI for predicting the efficacy of neoadjuvant chemotherapy in invasive breast cancer

Abstract

Objective: To establish and validate a radiomics nomogram based on contrast-enhanced (CE)-MRI for predicting the efficacy of neoadjuvant chemotherapy (NAC) in human epidermal growth factor receptor 2 (HER2)-positive breast cancer with non-mass enhancement (NME).

Methods: A cohort comprising 117 HER2-positive breast cancer patients showing NME on CE-MRI between January 2012 and December 2019 were retrospectively analysed in our study. Patients were classified as pathological complete respone (pCR) according to surgical specimens and axillary lymph nodes without invasive tumour cells. Clinicopathological data were recorded, and images were assessed by two radiologists. A total of 1130 radiomics features were extracted from the primary tumour and six radiomics features were selected by the maximal relevance and minimal redundancy and least absolute shrinkage and selection operator algorithms. Univariate logistic regression was used to screen meaningful clinical and imaging features. The rad-score and independent risk factors were incorporated to build a nomogram model. Calibration and receiver operator characteristic curves were used to confirm the performance of the nomogram in the training and testing cohorts. The clinical usefulness of the nomogram was evaluated by decision curve analysis.

Results: The difference in the rad-score between the pCR and non-pCR groups was significant in the training and testing cohorts (p < 0.01). The nomogram model showed good calibration and discrimination, with AUCs of 0.900 and 0.810 in the training and testing cohorts. Decision curve analysis indicated that the radiomics-based model was superior in terms of patient clinical benefit.

Conclusion: The MRI-based radiomics nomogram model could be used to pre-operatively predict the efficacy of NAC in HER2-positive breast cancer patients showing NME.

Advances in knowledge: HER2-positive breast cancer showing segmental enhancement on CE-MRI was more likely to achieve pCR after NAC than regional enhancement and diffuse enhancement.The MRI-based radiomics nomogram model could be used to pre-operatively predict the efficacy of NAC in HER2-positive breast cancer that showed NME.

Figure 1.

(a) A 51-year-old female showing segmental enhancement in the right breast; biopsy pathology: invasive breast cancer. (b) A 49-year-old female showing regional enhancement in the left breast; biopsy pathology: invasive breast cancer. (c) A 31-year-old female showing diffuse enhancement in the right breast; biopsy pathology: invasive breast cancer IHC: ER negative, PR negative, HER2 positive; surgical pathology: non-pCR. IHC, immunohistochemistry; pCR, pathological complete response; HER2, human epidermal growth factor receptor 2.

Figure 2.

Radiomics feature selection by using mRMR and LASSO. (a) Tuning parameter (λ) selection in the LASSO model used tenfold cross-validation via minimum criteria. The LASSO coefficient profiles of the 30 radiomics features. (b) A coefficient profile plot was produced against the log (I) sequence, and six features were chosen. LASSO, least absolute shrinkage and selection operator; mRMR, maximal relevance and minimal redundancy.

Figure 3.

The developed imaging-radiomics nomogram for predicting the efficacy of NAC in HER2-positive breast cancer that showed non-mass enhancement. For distribution, 1 represents segmental enhancement, 2 represents regional enhancement, and 3 represents diffuse enhancement. By calculating the scores of each point and locating it on the total score scale, the estimated probability of pCR can be assessed. HER2, human epidermal growth factor receptor 2; NAC, neoadjuvant chemotherapy; pCR, pathological complete response.

Figure 4.

Calibration curves of the prediction nomogram in the training (a) and testing (b) sets. Calibration curves depict the calibration of the nomogram according to the agreement between the probability of the incidence of pCR and the actual observation. The black line represents the ideal estimation, and the red line represents the apparent prediction of the nomogram. The closer the red line is to the ideal black line, the better the prediction ability of the nomogram. pCR, pathological complete response

Figure 5.

DCAs for the imaging and combined imaging-radiomics models. The grey line represents the assumption that all patients have breast cancer. The black line represents the assumption that none of the patients have breast cancer. The blue line represents the imaging model. The red line represents the imaging-radiomics model. Across the various threshold probabilities, the imaging-radiomics curve shows a maximised net benefit compared with the imaging model for the individual performance. DCA, decision-curve analysis.