Details

Autor(en) / Beteiligte

• Abdel-Fatah, Tarek Ma
• Ball, Graham
• Chen, Xin
• Mehaisi, Dalia
• Giannotti, Elisabetta
• Auer, Dorothee
• Vadakekolathu, Jayakumar
• Li, Ruizhe
• Pockley, Graham
• Chan, Stephen

Titel

Abstract P1-08-19: Utilising artificial intelligence (AI) for analysing multiplex genomic and magnetic resonance imaging (MRI) data to develop multimodality predictive system for personalised neoadjuvant treatment of breast cancer (BC)

Ist Teil von

• Cancer research (Chicago, Ill.), 2022-02, Vol.82 (4_Supplement), p.P1-P1-08-19

Erscheinungsjahr

2022

Link zum Volltext

Quelle

EZB Free E-Journals

Beschreibungen/Notizen

• Abstract Background: No effective tool is available to predict response to neoadjuvant chemotherapy (NACT). Aim: to enable an accurate prediction of the treatment response for BC patients, with the aid of machine learning based analysis of tumour histopathological, molecular & imaging features before NACT. Methods: The baseline biopsies were immunohistochemistry (IHC) stained for ER, PR, HER2, Ki67 & SPAG5 and genomically analysed. To build up a genomic profiler & multiplex analysis, a tailored mRNA code set of 450 genes was designed using NanoString nCounter Flex Analysis System. In addition, samples have been analysed for currently used prognostic & predictive genomic panels including: Oncotype DX [Genomic Health], MammaPrint [Agendia], Prosigna [Nanostring Technologies] & MD Anderson Genomic Chemo Sensitivity Predictor. The imaging features were extracted from MRI using radiomics measurements and deep learning methods. Our predictive models were trained in a local cohort of 400 BC (300 HER2 negative (HER2-) received 8 cycles of anthracycline & Taxane & 100 HER2 positive (HER2+) received Trastuzumab, anthracycline &Taxane) & was validated in 150 patients from local clinical trial patients. I-SPY-1 clinal trial genomic & imaging data were used as an external validation. The pathological complete response (pCR) defined as absence of tumor cells in both breast & lymph node was used as endpoint. Results: Compared to HER- BC, HER2+ BC had higher expression of GRB7, ERBB2, SPAG5, FGRF4 & CDC6. While HER2 negative BC showed higher level of CCND1, CDK6, FOXM1, FOXC1, MAI, MYC, ANP32E, SFRP1, ACTR3B, RAD51AP1, MCM3, LIN9, CDCA7 & PMAIP1 (Table 1). In both HER2 positive and negative BC; the reduction of >70% in tumour volume (OR (95% CI)=1.72 (1.35-2.7, p<0.0001) & 1.87 (1.44-2.42, p<0.0001);respectively), high mitotic count (OR (95% CI)=2.6 (1.73-3.91), p<0.0001) & 1.98 (1.49-2.65), p<0.000);respectively), ER- (OR (95% CI)=0.51 (0.41-0.64); p<0.0001 & 0.53 (0.42-0.68; p<0.0001; respectively, IHC), PR- (OR (95% CI)=0.67 (0.57-0.80: p<0.0001 & 0.62 (0.51-0.77; p<0.0001; respectively, IHC), SPAG5+ (OR (95% CI)=2.36 ( 1.77-3.15; p<0.0001) & 2.15 (1.51-3.06); p=<0.0001); respectively, IHC) and over-expressions of ERBB2 (OR (95% CI)=3.51 (2.26-5.45) & 3.32 (2.15-5.10) ; ps<0.0001, respectively), EGFR (OR (95% CI)=1.43. (1.01-2.03; p=0.004) & 1.48 (1.16-1.89); p=0.002; respectively), GRB7 (OR (95% CI)= 3.78 (2.52-5.64) & 3.56 (2.39-5.29); respectively, ps<0.0001), and CDC6 (OR (95% CI)=2.19 1.65-2.90; p=0.0002 & 2.1 (1.60-2.77); p=0.0005; respectively) were associated with higher pCR after receiving NACT. In HER2+ BC, the differential expression of DLC1, EBB4, RUNDC1, ARHGEF9 & PIK3CD were associated with higher response to Trastuzumab (ps<0.01).In HER2- BC: the intensity of histogram skewness in MRI (p=0.03) and low expression of CXXC5, AR, TGFB3, TACC3, TUBA4A, AGR2, ESR1, TP73 and BAG1 were associated with high response to chemotherapy (ps<0.01). Integrated models were developed for predicting response to NACT of HER2- BC (AUC (95% CI)= 0.83 (0.76-0.91; p<0.0001) & to chemotherapy plus Trastuzumab in HER2+ BC (AUC (95% CI) = 0.79 (0.67-0.91; p<0.0001). Conclusions: A predictive model was developed for HER2+ and another for HER2- BC with >= 80% accuracy prediction of pCR. AI & multiplex technology could enable robust biomarker discovery. ORLower 95% CIUpper 95% CIAdjusted pGRB7-mRNA9.986.7714.706.76E-24ERBB2-mRNA7.765.3011.401.12E-20FOXC1-mRNA0.100.060.166.44E-19MIA-mRNA0.180.110.282.63E-12ANP32E-mRNA0.500.410.622.30E-09MYC-mRNA0.520.410.678.82E-07SFRP1-mRNA0.310.190.491.65E-06CD99-mRNA1.481.261.732.73E-06SPAG5-mRNA1.891.452.474.68E-06PMAIP1-mRNA0.510.390.685.03E-06CDK6-mRNA0.520.390.699.09E-06FGFR4-mRNA2.621.733.979.21E-06CDCA7-mRNA0.400.270.591.13E-05LIN9-mRNA0.610.490.761.25E-05CDC6-mRNA1.941.452.591.44E-05MCM3-mRNA0.660.550.801.55E-05RAD51AP1-mRNA0.570.450.731.58E-05CCND1-mRNA0.450.310.641.93E-05FOXM1-mRNA0.510.380.681.97E-05 Citation Format: Tarek Ma Abdel-Fatah, Graham Ball, Xin Chen, Dalia Mehaisi, Elisabetta Giannotti, Dorothee Auer, Jayakumar Vadakekolathu, Ruizhe Li, Graham Pockley, Stephen Chan. Utilising artificial intelligence (AI) for analysing multiplex genomic and magnetic resonance imaging (MRI) data to develop multimodality predictive system for personalised neoadjuvant treatment of breast cancer (BC) [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P1-08-19.