Interobserver Variability in Evaluation of Physical Breast Phantoms

Employing 3D printing for constructing in-house physical phantoms offers a flexible and cost-effective approach, enabling the generation of diverse physical models in terms of both shape and content, tailored to the specific radiologic task. Following the production and imaging of the models, it is essential to ensure optimal evaluation by a radiologist. The synthetic image deviates from the clinical image dataset, presenting a challenge for radiological evaluation. Just as in clinical practice, training the radiologist to detect specific objects within the phantoms is crucial. The objective of this study is to assess the ability of 10 radiologists, who lack expertise in artificial and model imaging, in detecting breast lesions and assessing the realism of breast components. For this purpose, six physical phantoms, which differ in level of structural complexity, are used. A survey of six questions is prepared in respect to the realism of the phantoms’ x-ray images. The findings revealed that phantoms with air demonstrated greater contrast between lesions and other breast structures received higher ratings from the radiologists. Furthermore, the interobserver study indicated a low reliability among radiologists, attributed to the fact that they are not familiar with analyzing and reporting non-clinical images, especially those derived from physical phantoms. Radiologists engaged in the domain of in silico medicine should undergo advanced training to proficiently evaluate artificial and other non-clinical images.