Assignment of the Biological Identity of Breast Masses in Contrast-enhanced Spectral Computed Tomography Examinations...

Focal breast lesions are observed in up to 5.8% of CT examinations performed in female patients for a wide variety of indications not affecting the breast. To simplify and standardize the further procedure in the case of breast masses visualized by computed tomography (CT), an easy and robust diagnostic approach in assigning surely benign findings, uncertain findings and probably malignant findings is warranted.

We aimed (1) to determine the diagnostic gain of quantitative measurements of the iodine content in breast masses visualized in contrast-enhanced dual energy (DE) CT examinations additional to the assessment of qualitative image features; and (2) to generate an interactive hierarchical decision tree allowing to assess the individual probability of benign or malignant findings.

All contrast-enhanced DECT examinations performed in adult females from January 2018 to December 2020 were retrospectively screened for solid breast masses. Subsequently, the DECT images with confirmed breast masses were evaluated independently by 3 radiologists experienced in breast imaging using a structured questionnaire. First, the image characteristics of the ACR-BIRADS atlas on MRI were queried, and an assessment of the probability of benignancy or malignancy was requested. Second, the iodine content in the breast masses was interactively measured, and the readers were asked to provide a further assessment of the biological identity. Third, a decision tree was developed using a nonparametric supervised (machine) learning algorithm.

During the survey period, 5.591 contrast-enhanced DECT examinations of the chest were carried out in adult female patients. Six hundred eighty-seven DECTs (12.3%) displayed breast masses. The study group comprised 415 female patients (7.4%) with a benign or malignant breast mass confirmed by histology (155, 37.4%), a constant CT-course of ≥2 years without interfering therapy (202, 48.8%) and/or noninvasive breast imaging (57, 13.8%). Neither in the first step nor in the second step of the evaluations a cancer confirmed in the reference standard was classified as certainly benign. Subgroup analyses showed that the combined qualitative and quantitative approach was diagnostically superior compared with the sole descriptor-based assessments in 757 assessments of DECT evaluations with ACR 1+2 breast densities (AUC = 0.961 vs 0.924, lower bound 0.940 vs 0.893, upper bounds 0.982 vs 0.956, P = 0.001) and for the 541 assessments of DECT examinations visualizing masses in denser ACR 3+4 parenchyma (AUC = 0.972 vs 0.948, lower bound 0.953 vs 0.923, upper bound 0.991 vs 0.974, P = 0.043). Including iodine content into the descriptive image evaluations proved to be superior to the solely descriptor-based assessments in 456 assessments of lesions with >5 to 10 mm diameter and in 370 assessments of lesions with >10 to 20 mm ( P = 0.002, 0.002 and 0.018, respectively). Mass contour proved to be the first hierarchical decision level of the interactive classification tree for the differentiation of benign and malignant lesions developed in step 3. In the second hierarchical level, automatically generated classifications of the iodine content allowed to generate 3 diagnostic predictors analogous to the BIRADS classification with a depth of 2 ramifications.

Information on the iodine content of breast masses detected by DECT increases diagnostic certainty in the assessment of the biological significance of the lesions. The presented nonparametric decision tree may be helpful in interactively combining semantic and quantitative image data in clinical routine diagnostics of breast masses in spectral CT examinations.