Evaluation Of Computer-Aided Detection System In Mammography

Objective This study is:to verify retrospectively the validity of the computer-aided detection system MammoCAD2.0 in detecting characteristics of breast cancer by mammography,using breast cancer cases conformed histologically to estimate the sensitivity of CAD and using normal cases to estimate the false positive rate of CAD;to research effect of breast density on the performance of CAD; to investigate the actions of CAD for inexperience radiologists.Materials and Methods There were totally 617 cases including 415 breast cancer cases and 202 normal cases in this study.All breast cancers were confirmed by operation and all normal cases were proved by clinical date,sonography and mammography.All mammograms were evaluated for breast density and assigned a BI-RADS breast density score of 1 through 4 by two experienced radiologists.Images were reviewed and classified by two experienced radiologists.Mammograms with BI-RADS 1 density or BI-RADS 2 density were categorized as nondense. Mammograms with BI-RADS 3 density or BI-RADS 4 density were categorized as dense.The mammograms were detected with CAD,the sensitivity of breast cancers, MC and MA were evaluated respectively.Percentage is used to calculate the sensitivity of breast cancer characteristics,and compute the 95 percentage confidence interval.The design of the study used sensitivity of breast cancer and false positive rate to assess the performance of MammoCAD2.0.The total number of false-positive marks per case was determined by adding the number of microcalcification(MC), mass(MA)and lymph node(LN)marks in each normal case.Chi-square analysis was used to compare the sensitivity of breast cancer,MC and MA in normal dense and nondense.The number of CAD marks in normal dense and nondense mammograms was compared using chi-square analysis.Statistical analysis showing p values of less than 0.05 were considered statistically significant.226 patients with breast cancer conformed histologically were selected retrospectively in this study.Four methods were used in detecting breast cancer:CAD alone,radiologist with and without the CAD and experienced radiologist alone.CAD was used.The radiologists detect on high resolution displayer.Chi-square test was applied to data's analysis.Statistical analysis showing p values of less than 0.05 were considered statistically significant.Result Of the 415 breast cancer cases,289 cases manifest as MA,295 cases manifest as MC,108 cases were LN positive mammographically.There were 692 cases totally.CAD detected 271 MC cancer cases,249 MA cancer cases and 91 LN positive cases.Of the 202 normal cases,there were 404 mammograms.CAD make 319 marks,including 95 MC markers,184 MA markers and 40 LN marks.The average number of FALSE POSITIVE marks per case was 0.889,with 0.235 MC markers,0.455 MA markers and 0.099 LN marks per case.Of the 415 cancer cases, 88 were classified as BI-RADS 1,143 were classified as BI-RADS 2,157 were classified as BI-RADS 3 and 27 were classified as BI-RADS 4.The CAD detects 221 of 231 nonsense cases,165of 184 dense cases,including MC 140 cases and MA170 cases,MC 131 cases and MA 79cases respectively.The CAD is more sensitivity in detecting MC than MA statistically(P=0.000).The CAD detects more in nondense than in dense breasts(P=0.000),especially for the cases manifested as MA(P=0.001). But for the cancers manifested as MC mammographically,there was no statistically significant difference in the CAD performance in nondense versus dense breasts (P=0.891).Of the 202 normal cases,study radiologists classified 26 as BI-RADS 1,55 as BI-RADS 2,87 as BI-RADS 3 and 34 as BI-RADS 4.There were 81 nonsense and 121 dense breast cases detected by CAD.In nondense cases,20 cases have three or more FALSE POSITIVE marks,and 19 have three or more in dense cases.There was no statistically significant difference in nondense than in dense breast cases(P=0.113).Of the 226 cancer cases,there are totally 216 MAes cases and 86 MC cases.CAD detect 86.73%of breast cancer cases,including 83.33%MA cases and 84.88%MC cases.The inexperienced radiologist detects 89.82%cancer cases,with 85.65%MA cases and 80.23%MC cases.With the CAD cues,the inexperienced radiologist detects 97.35%cancer cases,with 96.76%MA cases and 98.84%MC cases.The experienced radiologist detects 99.12%cancer cases,with 99.07%MA cases and 100%MC cases.The sensitivity of breast-cancer,MA,MC was significantly increased after implementation of CAD,respectively.The detection rate of MC increased more than that of MA.There were no significant differences between the inexperienced radiologist with the CAD software and the experienced radiologist for both MA and MC.CAD makes only one MC false positive mark on the 2 missed cancers.It does not hint the experienced radiologists.CAD makes 370 false positive marks,including 188 MA markers and 182 MC markers.The average number of false positive marks per case was 0.82,with 0.40 MC markers and 0.42 MA markers per case.Conclusion CAD can mark the doubtful areas effectively on the mammography to cue the radiologists.The false positive rate is acceptable.There is statistically significant difference in breast cancer detection in dense and nondense breasts, especially for the cancer manifesting as MA.The detection of breast cancer manifesting as MC is not impacted by breast density.The false positive rate is not impacted by breast density.No statistically significant differences were seen in detecting breast-cancer,MA and MC between CAD and the inexperienced radiologist alone.The detection rates of breast-cancer,MA,MC for the inexperienced were significantly increased after implementation,respectively.There were no significant differences statistically in the detection of breast-cancer,MC and MA between the inexperienced radiologist with CAD and the experienced radiologist.There were significant differences between the inexperienced and the experienced radiologist in the detection of breast cancer,MA and MC.CAD isn't a replacement for the inexperienced radiologist,and it serves as an adjunct tool to assist the inexperienced radiologist in detecting breast cancer.