JMIR Publications

ABSTRACT

Background:

Computer-aided Diagnosis (CAD) may be an easy tool for radiologists to use on breast lesions in ultrasonography(US). Previous studies demonstrated that CAD can help to reduce the incidence of missed diagnosis for radiologists. However, the optimal method of diagnostic planes to assess tumor has not been mentioned.

Objective:

To explore the diagnostic performance of CAD in quadri-planes(QP) method on detecting breast cancer.

Methods:

We prospectively enroll 331 consecutive patients (age, mean[SD], 43.88[12.10] years; range, 17-70 years), 512 lesions (mean[SD], 1.85[1.19], range,0.26-9.5cm; 200 with malignant and 312 with benign), with breast mass as prominent symptom. Blindly to each other, one novice radiologist and one experienced radiologist are both assigned to read US images of lesions without CAD, according to the fifth edition of the American College of Radiology Breast Imaging Reporting and Data System lexicon and final assessment categories. The results of CAD were in dichotomized form: possibly benign and possibly malignant, and final assessments are composed by radiologist and CAD. We then compare the diagnostic performance of the readers in without-CAD mode, CAD in QP method, and readers combined with CAD in QP method.

Results:

The area under the receiver operating characteristic curve(AUC) of CAD in QP method is 0.90, based on the Youden index(0.68), the cutoff value of positive planes are greater or equal to 2.5, we choose two positive of four planes as threshold, the AUC of QP is significantly higher than CP method (0.84 vs 0.76;P＜.001]). The AUC of the novice radiologist is significantly improved from .732(without-CAD) to 0.843 (in QP; P＜.001). For the experienced reader, those are mildly improved from 0.845(without-CAD) to 0.87(in QP; P=.15). The consistency between experienced reader and novice reader in QP method is good (kappa value 0.63). The sensitivity, specificity, PPV, NPV, and accuracy of CAD in QP method are 87.50%, 81.41%, 75.11%, 91.04%, and 83.79%. Combining with CAD, sensitivity and NPV of the novice reader significantly improved from 60.0% to 79.0% (P=.004), 76.81% to 86.71%( P=.07), respectively, while, specificity, PPV and accuracy are mildly improved from 84.94% to 87.82%(P =.53), 71.86% to 80.61% (P =.13), and 75.20% to 84.38% (P =.12), respectively. For the experienced reader, sensitivity also improved significantly from 76.00% to 87% (P=.045), others with mildly changed without significance, specificity 92.95% down to 87.18%(P =.16), PPV 87.36% down to 81.31%( P =.25), NPV 85.80% up to 91%(P =.27), accuracy 86.33% to 87%(P =.84).

Conclusions:

The QP method for CAD has a better precision, with good consistency with the experienced radiologist, which can significantly improve the diagnostic performance of the novice radiologist, having an important application value in clinical diagnosis of breast cancer on ultrasonography. Clinical Trial: ChiCTR1800019649 (http://www.chictr.org.cn)