Artificial Intelligence: A Bayesian, Probability-Based
Diagnostic & Management Algorithm for Breast
Cancer
Dale Srinivas* Anu Singh
Twin Island University , Bushlot , Guyana Email: krishna28us@[Link]
Abstract
Background: Breast cancer evaluation spans screening, diagnostic imaging, and tissue
confirmation. Decisions are often qualitative. Objective: Present a Bayesian
probability-based algorithm that starts with a prior (risk model/BI-RADS) and applies
sequential updates from multimodal imaging and biopsy to guide action thresholds.
Methods: Standard breast imaging pathways (mammography/tomosynthesis, ultrasound,
MRI) are mapped to Bayesian updates; BI-RADS categories provide priors. Orthogonal
confirmation uses image-guided core biopsy. Management links receptor status and stage to
initial therapy.
Conclusion: The framework quantifies uncertainty and aligns imaging/biopsy results with
clear actions, improving documentation and MDT communication.
1) Define the Clinical Question
Does this patient have current breast malignancy, and if so, what is the immediate next step
(biopsy, further imaging, or surveillance) based on post-test probability?
2) Establish a Prior Probability
Set the prior using context (screen-detected vs symptomatic), risk models (e.g., Gail,
Tyrer-Cuzick), density, and the initial imaging impression. For diagnostics, the BI-RADS
category provides an approximate probability band: 3 (<2%), 4A (2–10%), 4B (10–50%),
4C (50–95%), 5 (>95%).
3) Sequential Bayesian Updates
A. Mammography/tomosynthesis: spiculation, architectural distortion, suspicious
calcifications increase probability.
B. Ultrasound: irregular/heterogeneous hypoechoic mass with posterior shadowing,
non-circumscribed margins increases probability; simple cyst or typical fibroadenoma
patterns decrease it.
C. MRI (problem-solving/high risk): early enhancement with washout kinetics, irregular
margins, and restricted diffusion increase probability.
D. Risk recalculation: BI-RADS may be upgraded/downgraded after adjunct imaging,
functioning as a formal Bayesian step.
4) Orthogonal Confirmation & Concordance
Perform image-guided core-needle biopsy when post-test probability crosses the biopsy
threshold (generally BI-RADS 4/5). Ensure radiology-pathology concordance; discordant
benign results warrant re-biopsy or surgical excision.
5) Decision Thresholds (Probability → Action)
• Very high (≥95% or BI-RADS 5): treat as malignant pending core; expedite tissue
diagnosis and staging.
• Intermediate (≈10–65% or BI-RADS 4A/4B): biopsy or additional orthogonal imaging
based on feasibility and patient preference.
• Very low (≤2% or BI-RADS 3): short-interval follow-up (e.g., 6 months).
6) Worked Example (Illustrative)
7) Staging & Initial Management (Once Malignant)
• Stage with clinical exam ± targeted axillary ultrasound/biopsy; cross-sectional imaging as
indicated.
• Surgical planning: breast-conserving surgery (BCS) vs mastectomy; sentinel node biopsy
(± axillary dissection if indicated).
• Systemic therapy guided by ER/PR/HER2 and, where appropriate, genomic assays (e.g.,
recurrence risk testing) to decide on chemotherapy.
• Neoadjuvant therapy for selected HER2-positive or triple-negative cancers, and
downstaging for BCS.
8) Documentation Essentials
Record prior anchors (risk model, BI-RADS, density), sequential imaging updates, post-test
probability/threshold crossed, biopsy route and results, receptor status, stage, MDT plan,
and shared decisions.
Figures
Notes
This probability-based framework is illustrative and should be adapted to local protocols
and resource availability. Exact likelihood ratios vary by technology, reader performance,
and population risk.
Supplemental Panel: Screening Risk (Gail/Tyrer-Cuzick Inputs & Imaging
Implications)
Supplemental Panel: BI-RADS Quick Reference
Supplemental Panel: Treatment Branch by Receptor Subtype