Tumoral mass classification by specialists and the CAD scheme

Abstract

This work consists of the validation of a Computer-Aided Diagnostic (CAD) scheme for mammographic images that to be adopted in the digital mammographic image interpretation training system for radiology residents at UNIFESP. It is a refinement phase of a CAD, developed by the LAPIMO (Laboratory of Medical and Odontology Images Processing and Analysis) group from EESC / USP, São Carlos. As a part of the CAD validation, digital mammograms were used, with double reading by specialists, all with pathological confirmation of lesions. 102 malignant lesions were used and the specialists responses were compared to those given by the CAD, both of them according to BI-RADS standards. To evaluate the CAD performance, the BIRADS categories 0, 4 and 5 were considered as positive lesions and the categories 1, 2 and 3 as negative. Comparing the CAD readings to those of the specialists, we obtained the following results: true-positive (TP) - 0.84 and 0.74, respectively; false-negative (FN) - 0.16 and 0.26, respectively. The agreement between the CAD readings and those of the specialists was of approximately 61%. The CAD and specialists FN readings disagreed in approximately 90% of the cases. At this validation phase, the CAD’s contribution to the FN reduction as a second opinion in the diagnosis became clear.