Project Summary
Image-guided core biopsy is the standard of care for the diagnosis of a suspicious finding in the breast. Unfortunately, the assessment of malignancy risk following breast core biopsy is imperfect and biopsies can be “non-definitive” in 5-15% of cases [1-7]. A non-definitive result means the chance of malignancy remains high due to possible sampling error (i.e. obtained biopsy is not representative of the suspicious finding) for which surgical excisional biopsy or aggressive radiologic follow-up is proposed. “Non-definitive” biopsies may therefore result in missed breast cancers (false negatives) and unnecessary interventions (false positives). Recent statistics show that one woman dies of breast cancer every 13 minutes in the U.S. and in 2012, an estimated number of 39.510 women (15% of all deaths) and 410 men in the U.S. are expected to die from breast cancer. In Portugal, are detected annually approximately 4500 new cases which lead to the death of almost 1500 women. This important problem is emblematic of a plethora of clinical situations where rigorous and accurate risk estimation of rare events using graphical models have the potential to enhance clinician decision-making and provide the opportunity for shared decision making with patients in order to personalize and strategically target health care interventions.
The goal of this research is to integrate physician expertise and machine learning into graphical models that will accurately estimate breast cancer risk after image-guided breast biopsy, in order to address the problem of biopsy sampling error.
It is proposed to develop a novel methodology for building decision support systems called Advice-Based-Learning (ABLe) that will iteratively combine heterogeneous clinical data using a probabilistic form of inductive logic programming (ILP) and expert physician knowledge to train accurate graphical models designed specifically for clinical translation.
Bibliographical References:
| [1] | Burbank F. Stereotactic breast biopsy: comparison of 14- and 11-gauge Mammotome probe performance and complication rates. Am Surg. 1997;63(11):988-95. |
| [2] | Philpotts LE, Shaheen NA, Carter D, Lange RC, Lee CH. Comparison of rebiopsy rates after stereotactic core needle biopsy of the breast with 11-gauge vacuum suction probe versus 14-gauge needle and automatic gun. AJR Am J Roentgenol. 1999;172(3):683-7. |
| [3] | Liberman L. Centennial dissertation. Percutaneous imaging-guided core breast biopsy: state of the art at the millennium. AJR Am J Roentgenol. 2000;174(5):1191-9. |
| [4] | Berg WA, Hruban RH, Kumar D, Singh HR, Brem RF, Gatewood OM. Lessons from mammographic histopathologic correlation of large-core needle breast biopsy. Radiographics. 1996;16(5):1111-30. |
| [5] | Liberman L, Drotman M, Morris EA, et al. Imaging-histologic discordance at percutaneous breast biopsy. Cancer. 2000;89(12):2538-46. |
| [6] | Siegmann KC, Wersebe A, Fischmann A, et al. [Stereotactic vacuum-assisted breast biopsy-success, histologic accuracy, patient acceptance and optimizing the BI-RADSTM-correlated indication]. Rofo. 2003;175(1):99-104. |
| [7] | Zuiani C, Mazzarella F, Londero V, Linda A, Puglisi F, Bazzocchi M. Stereotactic vacuum-assisted breast biopsy: results, follow-up and correlation with radiological suspicion. Radiol Med. 2007;112(2):304-17. |
Keywords:
Data MiningMachine Learning
Inductive Logic Programming
Probabilistic Graphical Models
