How to use the Kaiser score as a clinical decision rule for diagnosis in multiparametric breast MRI: a pictorial essay

Abstract

Due to its superior sensitivity, breast MRI (bMRI) has been established as an important additional diagnostic tool in the breast clinic and is used for screening in patients with an elevated risk for breast cancer. Breast MRI, however, is a complex tool, providing multiple images containing several contrasts. Thus, reading bMRI requires a structured approach. A lack of structure will increase the rate of false-positive findings and sacrifice most of the advantages of bMRI as additional work-up will be required. While the BI-RADS (Breast Imaging Reporting And Data System) lexicon is a major step toward standardised and structured reporting, it does not provide a clinical decision rule with which to guide diagnostic decisions. Such a clinical decision rule, however, is provided by the Kaiser score, which combines five independent diagnostic BI-RADS lexicon criteria (margins, SI-time curve type, internal enhancement and presence of oedema) in an intuitive flowchart. The resulting score provides probabilities of malignancy that can be used for evidence-based decision-making in the breast clinic. Notably, considerable benefits have been demonstrated for radiologists with initial and intermediate experience in bMRI. This pictorial essay is a practical guide to the application of the Kaiser score in the interpretation of breast MRI examinations.Teaching Points • bMRI requires standardisation of patient-management, protocols, and reading set-up. • Reading bMRI includes the assessment of breast parenchyma, associated findings, and lesions. • Diagnostic decisions should be made according to evidence-based clinical decision rules. • The evidence-based Kaiser score is applicable independent of bMRI protocol and scanner. • The Kaiser score provides high diagnostic accuracy with low inter-observer variability.

Implementation of machine learning into clinical breast MRI: Potential for objective and accurate decision-making in suspicious breast masses

“…In contrast to our approach, it aims for a quick visual assessment that omits any measurements, is based on a decision tree algorithm and has been validated in multiple centers. Studies verified its low interobserver variability and high diagnostic accuracy [7,29,35].…”

“…In addition, potential decision rules were evaluated. Such decision rules have been described as promising tools for breast MRI assessment [7,29], enabling exact diagnostic statements of either presence (rule-in) or absence of malignancy (rule-out). As a common principle, if sensitivity is high, a "negative" test result will rule out malignancy, and with a high specificity, a "positive" test result will rule in malignancy [30][31][32].…”

“…Hence, ML could assist the reader in making objective and accurate decisions in BI-RADS IV cases and thereby reduce the number of unnecessary biopsies by up to 31.3%. The literature reports a number of classification rules for breast MRI [7,16,29,[35][36][37], with most of them applying standard morphological criteria, and establishing a diagnosis by e.g. calculating simple sum scores [36,37].…”

“…ML, however, is able to detect interactions of higher complexity between imaging parameters, which translates into high accuracy [9,10]. Another approach different from sum score-based methods is the "Kaiser Score", which represents one of the best investigated classification algorithms in breast MRI [7,29,35]. In contrast to our approach, it aims for a quick visual assessment that omits any measurements, is based on a decision tree algorithm and has been validated in multiple centers.…”

“…Therefore, MRI is still regarded as an operator-dependent method, and considerable interobserver variability of the BI-RADS descriptors is well documented [5,6]. Moreover, BI-RADS is a formal lexicon and does not provide objective decision rules to integrate relevant information and provide a diagnosis [7,8]. Therefore, the final BI-RADS assessment rather represents a radiologist's subjective rating.…”