Studies in recent years have provided much information regarding sensitivity of CAD, as well as their utility. However, other studies such as the studies published by Fenton et al., have provided an opposing argument stating increased work-up rates, increased costs, and increase in radiologists’ time.[22
] In 2007, Fenton reported that sensitivity increased from 80.4% to 84.0% with new users as well as increased biopsy rate (19.7%).[22
] The study found that diagnostic specificity decreased from 90.2% to 87.2% after implementation of CAD, as did the positive predictive value, from 4.1% to 3.2%.This study reported that the change in cancer-detection rate was not significant and concluded that the use of CAD is associated with reduced accuracy of interpretation of screening mammograms. Fenton et al., in 2011, published a follow-up CAD study, again evaluating screen-film mammography. This study concluded that the use of CAD with film screen is associated with decreased specificity and no improvement in the detection rate of invasive cancer, although they did find increased sensitivity for ductal carcinoma in situ
] It is important to note that in both of the Fenton analyses, the facilities included were sites that had low volume, as well as radiologists who were inexperienced with using CAD. Both studies concluded that CAD use hinders accuracy of interpretation of screening mammograms. Our findings demonstrate that CAD does have value as it marked the carcinoma at the time of diagnosis 79% of the time, and when reviewing images from the year(s) prior, CAD marked the carcinoma 56% of the time. Burhenne and colleagues reviewed cancers and prior mammograms and found that CAD marked 77% of cases where the cancer was visible in retrospect.[1
] Our 2004 review found that CAD correctly marked 71% of actionable findings read as negative in previous years.[11
] As Burhenne's study points out, if ample attention is directed to the visible and actionable lesions, these lesions can be addressed at the time of presentation; thus, more cancers may be diagnosed earlier.
Lack of CAD interpretation training or education can often be a reason for overlooking true positive marks. Roehrig wrote that training radiologists in the use of CAD is more important than was originally thought.[24
] Traditionally a training session of 1 day has been performed; however, the author states there is evidence supporting the value of longer training. Astley and Gilbert, in a preliminary study, had a 7-week training program in the UK with promising results.[25
] Guerriero et al., in 2011, wrote that 4 days of CAD training was required for radiologists and additionally, those using CAD should be retrained every 3 years.[26
] Luo et al. showed a statistically significant difference in observer performance in utilizing CAD with mammography interpretation before and after training. The training consisted of three participants who read a pretest set of 80 (mixed) benign and malignant cases and after 4 weeks of training read the posttest set of cases. The authors concluded that CAD training influences perception, recognition, and interpretation of early breast cancer and CAD performance studies.[27
] We believe inadequate training may be a factor in some of the negative results reported in the literature that reveal reduction in sensitivity, increase recall rates, and increase biopsy rate. Training, as has been shown, may be valuable to the radiologist to understand the technology and thus the marks. The radiologist may acknowledge CAD marks but ultimately ignore them due to concentration in another area of the breast. The problem with this scenario may be a lack of trust of the marks and/or multiple false-positive marks. Research on interactive CAD technology is ongoing. One interactive method would allow the radiologist to click on an area of concern and only this area would display potential marks. This may provide a way to minimize false–positive marks and reduce the distraction of multiple marks, which can potentially lead to cancers being overlooked.[28
An interesting observation in our review was that there were no lesions with pure calcifications included in our study cohort. This leads us to believe that we are detecting and diagnosing calcifications. FFDM has helped identify tiny clusters of microcalcifications that may otherwise not be identified on routine screening mammography. Improvement in detecting calcifications has been reported previously.[29
] A high level of trust exists with CAD marking calcifications, so radiologists are acting on the CAD prompts when it does mark these lesions. This is supported by the literature, which has demonstrated the sensitivity of CAD for detecting malignant microcalcifications to be as high as 99%.[1
] Our review of evaluating CAD marking invasive lobular carcinoma found that CAD marked 100% of calcifications.[32
] This leads to the question, why are masses more frequently missed? The sensitivity of malignant masses has been reported at 75-89%.[1
] In this review, 59% (n
= 27/46) of lesions were marked on just one view. Radiologists may not trust the marks when just in one view; however, this study demonstrates that attention needs to be paid even in this scenario, as we found that CAD marked a more significant portion on the CC view (32%, n
= 11/34) compared with 21% (n
= 7/24), the MLO view. This is in agreement with our prior publication which found that CAD marked a majority of lesions in the CC view.[20
This review showed that CAD does have the ability to mark lesions in dense breast tissue. In the year of diagnosis, CAD marked 79% in heterogeneously dense or extremely dense tissue. CAD in prior year(s) marked 80% in heterogeneously dense or extremely dense breasts. Lesion size of those detected and not detected by CAD were very similar, with the average being 14 and 13 mm, respectively. A study by Brem and colleagues also found no significant difference in performance based on cancer size.[18
A study limitation is that several different versions of CAD were used throughout the study period. This could have led to variations in what CAD did and did not mark. Additionally, some of the prior studies were film screen and having all digital priors would have minimized technical variables, although the priors were all digitized for comparison at the time of diagnosis and for the retrospective review. Additionally, we had a mixture of diagnostic and screening patients at the time of presentation and cancer diagnosis. However, we view our diagnostic and screening patient images and CAD similarly.
In conclusion, this review revealed that CAD marked the breast carcinoma on the mammogram at the time of diagnosis 79% of the time (the radiologists detected 100% of the cancers) and CAD marked the visible cancer 56% of the time, in the prior-year(s) mammograms (the radiologists did not detect). Even though CAD marked the lesion of interest on the prior-year studies, radiologists continue to disregard the CAD marks. As no calcifications were missed in this cohort, we were able to presume that CAD is marking calcifications, and the radiologist is accurately working up these cases. Our review demonstrated that masses are being marked by CAD, even in dense breast tissue, and are invasive carcinomas of significant size. We found that CAD is marking a significant portion of lesions on only the CC view; potentially this is an indicator to radiologists to be especially vigilant when a lesion is marked on this view.