| Περιγραφή: |
Digital 3D mammography is currently the most utilized screening tool used to identify early-stage breast cancer. Mammograms are scored using the BI-RADS categories (Breast Imaging Reporting and Data System). Categories IV and V mammograms are associated with a significant risk of breast cancer, mandating a biopsy and a pathologic diagnosis. Image guided percutaneous needle biopsy diagnoses cancer in about 25% of these BIRADS IV, V abnormalities emphasizing the lack of specificity and sensitivity of 3d Digital Mammography in diagnosing early-stage breast cancer. The poor positive predictive score subjects 75% of these patients with benign radiographic abnormalities to the potential anxiety, cost and harms of an unnecessary tissue biopsy. We studied 150 patients with BIRADS IV, and V mammographic abnormalities with serum, plasma and saliva obtained just prior to image guided biopsy. Hydrogel affinity nanoparticles were used to harvest and concentrate low abundance proteins, and the proteins were then identified by mass spectrometry. Candidate peptides were identified, and these markers were verified in blinded confirmation. We developed and validated a statistical model to differentiate between cancer and non-cancer cases using a combination of proteomic and microbial biomarkers. Biomarkers were identified through bivariate analysis to find those with significant differences in abundance between cancer cases and controls. Additionally, lasso feature selection was employed to pinpoint the biomarkers most strongly associated with the cancer outcome. The model was constructed by comparing three distinct models using maximum likelihood estimates and Akaike information criterion, with the most parsimonious model selected for final analysis. K-fold cross-validation was used to ensure the model’s robustness, and a receiver operating characteristic (ROC) curve analysis yielded an area under the curve (AUC) of 0.82, indicating strong overall predictive accuracy. To determine risk categories, we identified thresholds using spline analysis and calibration models. Splines were applied to the predicted probabilities to detect significant changes in cancer likelihood, and calibration curves were used to fine-tune the threshold. This process identified a 0.5 probability cut-off, which was used to categorize predictions into low and high likelihood of cancer. Confusion matrices were then employed to evaluate model performance at this threshold. The model demonstrated high sensitivity (92%), effectively identifying true positive cases, but had a lower specificity (45%), indicating a higher rate of false positives. Despite challenges such as handling factor levels and missing values, the model, which incorporated nine significant biomarkers (seven proteomes and two microbes), proved valuable for identifying high-risk patients. The model’s high sensitivity potentially makes it a useful tool to guide the clinical decision to biopsy a mammographic abnormality or follow it clinically with additional interval follow-up imaging. An accurate risk marker can potentially reduce unnecessary invasive biopsies. To enhance clinical utility, future work should focus on improving specificity to reduce false positives and validating the model with larger, more diverse datasets to increase its generalizability and practical application. Citation Format: Kurayi Mahachi, Wendy Pelton, Angela Toepp, Alessandra Luchini, Lance Liotta, Richard Hoefer. Improving the Specificity and Sensitivity of Digital 3D Mammography using serum, plasma and saliva derived proteomes and mathematical modeling [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P1-05-25. |