Hunt, Katie N.

PURPOSE Background parenchymal uptake (BPU), which describes the level of radiotracer uptake in normal fibroglandular tissue on molecular breast imaging (MBI), has been identified as a breast cancer risk factor. Our objective was to develop and validate a deep learning model using image convolution to automatically categorize BPU on MBI. METHODS MBI examinations obtained for clinical and research purposes from 2004 to 2015 were reviewed to classify the BPU pattern using a standardized five-category scale. Two expert radiologists provided interpretations that were used as the reference standard for modeling. The modeling consisted of training and validating a convolutional neural network to predict BPU. Model performance was summarized in data reserved to test the performance of the algorithm at the per-image and per-breast levels. RESULTS Training was performed on 24,639 images from 3,133 unique patients. The model performance on the withheld testing data (6,172 images; 786 patients) was evaluated. Using direct matching on the predicted classification resulted in an accuracy of 69.4% (95% CI, 67.4% to 71.3%), and if prediction within one category was considered, accuracy increased to 96.0% (95% CI, 95.2% to 96.7%). When considering the breast-level prediction of BPU, the accuracy remained strong, with 70.3% (95% CI, 68.0% to 72.6%) and 96.2% (95% CI, 95.3% to 97.2%) for the direct match and allowance for one category, respectively. CONCLUSION BPU provided a robust target for training a convolutional neural network. A validated computer algorithm will allow for objective, reproducible encoding of BPU to foster its integration into risk-stratification algorithms.

Key Clinical Message This case highlights the role of molecular breast imaging (MBI) in evaluating persistent clinical concerns after a negative diagnostic mammogram and ultrasound. MBI is especially useful in the diagnosis of invasive lobular carcinoma due to its occult nature on conventional imaging modalities.

6531 Background: Annual mammography is recommended to screen residual breast tissue for new cancers and recurrent disease after treatment for early stage breast cancer. This study aimed to assess mammography rates over time in breast cancer survivors. Methods: We used administrative claims data from a large U.S. commercial insurance database, OptumLabs, to retrospectively identify privately- and Medicare Advantage-insured women with operable breast cancer who had residual breast tissue after definitive breast surgery between 2006 and 2015. We required coverage for at least 13 months following surgery. For each subsequent 13-month time period, we only included women without a loss of coverage, bilateral mastectomy, metastatic breast cancer diagnosis, or non-breast cancer diagnosis. We calculated the proportion of patients who had a mammogram during each 13-month period following breast surgery. We used multivariable logistic regression to test for factors associated with mammography in the first 13 months. Results: The cohort included 26,011 women followed for a median of 2.9 years (IQR 1.9-4.6) after surgery; 63.1% were less than 65 years of age, and 74.4% were white. In their first year of follow-up, 86% underwent mammography, but by year 7, this decreased to 73%. Fewer than 1% underwent MRI instead of mammography. In multivariable analysis, mammograms were less likely during the first year after surgery among women aged < 50 years (odds ratio [OR], 0.7; 95% confidence interval [CI], 0.6 to 0.8), African Americans (OR, 0.7; 95% CI, 0.7 to 0.8), patients who underwent mastectomy (OR, 0.7; 95% CI, 0.6 to 0.7), and patients residing in the Western part of the country (OR, 0.9; 95% CI, 0.7 to 0.9). Those with 1-2 comorbidities were more likely (OR, 1.1; 95% CI 1.1-1.2) than those with none to have a mammogram during that period. Mammography use did not differ significantly by year of diagnosis (2006-2015). Conclusions: Even in an insured cohort, a substantial proportion of breast cancer survivors do not undergo annual surveillance mammography. Mammography use falls as the time from the early stage breast cancer diagnosis increases. Understanding factors associated with lack of mammographic screening may help improve survivorship care.

Abstract Background. We previously reported an alternative ESR1 somatic gain-of-function chromosomal translocation event in a patient presenting with aggressive, endocrine therapy resistant estrogen receptor (ER) positive disease, producing an in-frame fusion gene consisting of N-terminal ESR1 and the C-terminus of the Hippo pathway coactivator YAP1 (ESR1-YAP1). We recently identified another ESR1 fusion through RNA sequencing (RNA-seq) in advanced stage ER+ disease from a chest wall recurrence in a male patient that was refractory to multiple lines of treatment. Two examples of fusions discovered in primary breast cancer samples include ESR1 fused in-frame to C-terminal sequences from NOP2 (ESR1-NOP2), identified in a resistant cohort from a RNA-seq screen focused on 81 primary breast cancers from aromatase inhibitor clinical trials, and a second ESR1 fusion, fused in-frame to the entire coding sequence of POLH (ESR1-POLH), that was identified from RNA-seq analysis of 728 Cancer Genome Atlas breast samples. This current study extends our previous characterization of ESR1-YAP1 by comparing functional and pharmacological properties of these three additional ESR1 gene fusion events of both early stage and advanced breast cancers. Methods. In vitro and in vivo experiments were conducted to test ESR1 fusions to induce therapeutic resistance, and metastasis. The transcriptional and binding properties of each fusion was also examined. Pharmacological inhibition with Palbociclib, a cyclin-dependent kinase 4/6 inhibitor, was utilized to assess drug sensitivity in ESR1 fusion containing breast cancer cells and in a patient derived xenograft (PDX) model expressing ESR1-YAP1 (WHIM18). Results. The YAP1 and PCDH11x fusions conferred estrogen-independent and fulvestrant-resistant growth. Immunohistochemistry revealed significantly higher numbers of ER+ cells in lungs of mice xenografted with T47D cells expressing the YAP1 and PCDH11x fusions compared to YFP control, NOP2 and POLH fusions. Results from ChIP-seq and microarray studies suggest that these two fusions promote proliferation and metastasis through genomic action by binding estrogen response elements (ERE) and subsequent gene activation. We thereby define these fusions as “canonical” fusions compared to “non-canonical” NOP2 and POLH fusions, which demonstrated dramatically decreased genomic binding ability. The non-canonical fusions induced genes associated with basal-like breast cancer and promoted HER2, EGFR, and MAPK gene expression signatures in contrast to genes associated with cell cycle/proliferation induced by canonical fusions. The proliferative ability of canonical fusion-containing ER+ cells was inhibited by Palbociclib in a dose-dependent manner. In vivo WHIM18 tumors in mice fed with Palbociclib-containing chow demonstrated significantly reduced tumor volume, growth rate, and weight compared to tumors in mice on control chow. Conclusions. In-frame ERE activating canonical fusions occur in end-stage drug resistant advanced breast cancer and can be added to ESR1 point mutations as a class of recurrent somatic mutation that may cause acquired resistance. Growth induced by these fusions can be antagonized by Palbociclib and is potentially clinically helpful. Citation Format: Lei JT, Shao J, Zhang J, Iglesia M, Cao J, Chan DW, He X, Kosaka Y, Schmidt C, Matsunuma R, Haricharan S, Crowder R, Hoog J, Phommaly C, Goncalves R, Ramalho S, Lai W-C, Hampton O, Rogers A, Tobias E, Parikh P, Davies S, Ma C, Suman V, Hunt K, Watson M, Hoadley KA, Thompson A, Chen X, Perou CM, Creighton CJ, Maher C, Ellis MJ. Recurrent functionally diverse in-frame ESR1 gene fusions drive endocrine resistance in breast cancer [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr PD2-03.

205 Background: Central review is an important trial management and validation tool. Multisite trials require data systems to accommodate diverse image acquisition and review. Breast cancer is challenging as its imaging objects are not always easily shared across departments. Quality Assurance Review Center (QARC) has a diverse portfolio of facile data acquisition and powerful informatics support that meets these needs. We report on the role of QARC in three of the current breast cancer trials from the American College of Surgeons Oncology Group. Methods: QARC data management system includes secure network infrastructure and a validated relational operating database (MAX). QARC has data acquisition/imaging re-distribution expertise for real time response reviews, response measuring tools and corresponding data archive for secondary reviews. MAX includes query ability; records link to digital information. QARC underwent systematic information systems validation process for 21 CFR Part 11. DICOM/non-DICOM files are stored in the QARC PACS. On-site/remote reviewers use MAX to retrieve, view, annotate and save images. Data extracted is securely sent to partner statistical centers. Results: In Z1031, a neoadjuvant endocrine therapy trial, all of the mammographic exams are archived at QARC. For Z1071, a sentinel lymph node trial, pre- and post-treatment ultrasound (US) images of 321 patients were remotely reviewed. For Z1072, the cryoablation US, pre and post-cryoablation MRI for 40 of 47 cases were remotely archived, retrieved and transferred to investigator workstation for post-processing and review. The other 7 patients did not have evaluable imaging studies. In all three trials, the costs of shipping hardcopies of exams and travel to QARC for central review functions were eliminated. Conclusions: QARC data management systems provide diverse informatics supports for multi-institutional trials, ranging from archives of images, provide remote access and download of data, and central review. The vibrant informatics supports meet the growing needs of clinical trials. Future directions include radiation field review for breast cancer trials and the incorporation of pathology microarray analysis as DICOM objects.

PurposeThe purpose of this study was to perform a pilot evaluation of an integrated molecular breast imaging/ultrasound (MBI/US) system designed to enable, in real‐time, the registration of US to MBI and diagnostic evaluation of breast lesions detected on MBI.MethodsThe MBI/US system was constructed by modifying an existing dual‐head cadmium zinc telluride (CZT)‐based MBI gamma camera. The upper MBI detector head was replaced with a mesh panel, which allowed an ultrasound probe to access the breast. An optical tracking system was used to monitor the location of the ultrasound transducer, referenced to the MBI detector. The lesion depth at which ultrasound was targeted was estimated from analysis of previously acquired dual‐head MBI datasets. A software tool was developed to project the US field of view onto the current MBI image. Correlation of lesion location between both modalities with real‐time MBI/US scanning was confirmed in a breast phantom model and assessed in 12 patients with a breast lesion detected on MBI.ResultsCombined MBI/US scanning allowed for registration of lesions detected on US and MBI as validated in phantom experiments. In patient studies, successful registration was achieved in 8 of 12 (67%) patients, with complete registration achieved in seven and partial registration achieved in one patient. In 4 of 12 (37%) patients, lesion registration was not achieved, partially attributed to uncertainty in lesion depth estimates from MBI.ConclusionThe MBI/US system enabled successful registration of US to MBI in over half of patients studied in this pilot evaluation. Future studies are needed to determine if real‐time, registered US imaging of MBI‐detected lesions may obviate the need to proceed to more expensive procedures such as contrast‐enhanced breast MRI for diagnostic workup or biopsy of MBI findings.

Abstract Background: Neoadjuvant aromatase inhibitor (AI) therapy would become a more acceptable alternative to chemotherapy if there was a way to identify AI-resistant cases early for triage to alternate systemic treatment. We therefore conducted a Phase 2 trial of post-menopausal patients with clinical stage 2 or 3 ER+ (Allred Score 6 to 8) breast cancer who were re-biopsied 2 to 4 weeks after initiating neoadjuvant AI therapy. If the tumor Ki67 level was &gt; 10% (AI resistant) the patient was triaged to either chemotherapy or immediate surgery and if &lt; 10% (AI sensitive) the patient completed 16 to 18 weeks of neoadjuvant AI treatment. Co-primary endpoints were: 1) to confirm that the pathological complete response (pCR) rate to standard NCCN chemotherapy in the AI resistant population is in the range of 20% (similar to ER− disease). This required the observation of least 4/35 pCRs; 2) whether patients were willing to accept a recommendation of no chemotherapy when the preoperative endocrine prognostic index score was zero (PEPI-0) (pStage 1 or 2A, Ki67&lt;2.7% and ER+) since, according to the PEPI model, these patients have such a low relapse risk that chemotherapy is unlikely to be of value (Ellis, M JNCI 100:1380–8, 2008). Feasible was defined as &lt;10% PEPI-0 cases receiving adjuvant chemotherapy. Secondary endpoints included a determination of the distribution of intrinsic subtypes in the AI sensitive and resistant groups (using a research use only assay). Results: 51 patients (21%) had 2 to 4 week Ki67&gt;10%, of these 36 received chemotherapy per protocol (27) anthracycline (A) and taxane (T)-based, 7 T-based, 1 A-only based and 1 other (non-NCCN) regimen but only 2 patients had a pCR (5.5%). 194 patients (79%) had a Ki67 &lt;10%, of these 176 had PEPI score information. 65 had a PEPI-0 (37%), of which 61 (94%) reported a treatment plan that did not include adjuvant chemotherapy. The AI resistant group had a significantly higher frequency of baseline high-risk intrinsic subtypes (LumB: 26, HER2-E: 2 versus LumA: 7) at baseline than the AI sensitive group (LumB: 40, HER2-E: 0 versus LumA: 66) Fishers exact test p = 7.16E−6. Additional molecular data will be presented at the meeting. Conclusion. A Ki67 value &gt;10% at 2 to 4 weeks strongly enriches for high-risk molecular subtypes (mainly LumB) however standard neoadjuvant chemotherapy did not meet pre-assigned criteria for adequate Phase 2 activity in this group. Novel neoadjuvant approaches for AI resistant-disease defined by high on treatment Ki67 are therefore a high priority. Management of patients with excellent response to neoadjuvant endocrine therapy (PEPI-0) without adjuvant chemotherapy is feasible. The ALTERNATE trial will enroll over 2000 patients to establish a new standard of care for neoadjuvant treatment of ER+ HER2− disease based on these principles. Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr PD07-01.

Abstract Background: Neoadjuvant endocrine therapy (NET) trials serve to inform design and selection of agents for larger adjuvant endocrine trials. The neoadjuvant aromatase inhibitor (AI) trial ACOSOG Z1031 randomized the treatment of postmenopausal women with ER rich (Allred 6-8) clinical stage 2/3 breast cancer to either EXE, ANA and LET. The results are therefore of interest in the context of new data that compares these AIs in the adjuvant setting. The opportunity to develop predictive biomarker tests for NET sensitivity was also a key study objective in order guide future selection of patients for inclusion in studies that compare NET versus neoadjuvant chemotherapy (NCT). The value of a baseline PAM50 intrinsic subtype test for NET patient selection was therefore an embedded objective of the study. Methods: Of the 374 women who began study treatment on Z1031, pretreatment biopsy specimen for central IHC testing for ER and Ki67 was available on 304 patients. The PAM50 intrinsic subtype test (PMCID: 2667820) could be conducted on 170 baseline samples. The Preoperative Endocrine Prognostic Index (PEPI) an approach to predicting long term outcomes based on pathological stage and surgical specimen ER and Ki67 status was also conducted because patients with a PEPI of zero (T1/2 N0, ER+, Ki67&lt;=2.7%) have been found to have an excellent long term prognosis and therefore unlikely to benefit from chemotherapy (PMCID: 255670). Results: The degree of Ki67 suppression, the absolute level of Ki67 in the surgical specimen and the PEPI results were not significantly different by treatment arm (Kruskal-Wallis tests p &gt;0.05). Thirty Percent (95%CI: 19.6-42.9%) of women with pre-AI LumA subtype and 12% (95%CI: 6.4-20.0%) of women with pre-AI LumB subtype had a PEPI of zero predicting excellent prognosis. Multivariate logistic modeling found that the likelihood of a PEPI score of zero is 3.2 fold higher (95%CI: 1.43 7.12) among 66 women with a pre-AI LumA subtype compared to 100 women with a pre-AI LumB subtype. Additional baseline variables (clinical T and N stage, HER2 status, pre-AI Ki-67 level (&lt; = 10% vs. &gt;10%) were not found to modify the chance of a PEPI of zero. Baseline PAM50 Subtype versus PEPI Conclusion: The results of Z1031 show no differences in clinical or biologic response between LET, EXE, and ANA in the neoadjuvant setting; predicting that no difference in efficacy will be seen between these agents in adjuvant studies. Pre-AI LumA versus LumB status, defined by the PAM50 subtype, identifies patients who are more likely to be in the most favorable PEPI score following neoadjuvant AI therapy. Patients with pretreatment LumA status may, therefore, not be suitable for a randomized trial of NET versus NCT since approximately one third of these patients may be able to avoid chemotherapy altogether. Citation Information: Cancer Res 2010;70(24 Suppl):Abstract nr S1-2.