Diagnostic Performance of Point-of-Care Apparent Diffusion Coefficient Measures to Reduce Biopsy in Breast Lesions at MRI: Clinical Validation

Abstract

Background

The Eastern Cooperative Oncology Group–American College of Radiology Imaging Network Cancer Research Group multicenter A6702 trial identified an optimal apparent diffusion coefficient (ADC) cutoff to potentially reduce biopsies by 21% without affecting sensitivity. Whether this performance can be achieved in clinical settings has not yet been established.

Purpose

To validate the performance of point-of-care ADC measurements with the A6702 trial ADC cutoff for reducing unnecessary biopsies in lesions detected at breast MRI.

Materials and Methods

Consecutive breast MRI examinations performed from May 2015 to January 2019 at a single medical center and showing biopsy-confirmed Breast Imaging Reporting and Data System category 4 or 5 lesions, without ipsilateral cancer, were identified. Point-of-care lesion ADC measurements collected at clinical interpretation were retrospectively evaluated. MRI examinations included axial T2-weighted, diffusion-weighted, and dynamic contrast-enhanced sequences. Sensitivity and biopsy reduction rates were calculated by applying the A6702 optimal (ADC, 1.53 × 10⁻³ mm²/sec) and alternate conservative (1.68 × 10⁻³ mm²/sec) cutoffs. Lesion pathologic outcomes were the reference standard. To assess reproducibility, one radiologist repeated ADC measurements, and agreement was summarized using the intraclass correlation coefficient.

Results

A total of 240 lesions in 201 women (mean age, 49 years ± 13 [SD]) with pathologic outcomes (63 malignant and 177 benign) were included. Applying the optimal ADC cutoff produced an overall biopsy reduction rate of 15.8% (38 of 240 lesions [95% CI: 11.2, 20.9]), with a sensitivity of 92.1% (58 of 63 lesions [95% CI: 82.4, 97.4]; sensitivity was 97.2% [35 of 36 lesions] [95% CI: 82.7, 99.6] for invasive cancers). Results were similar for screening versus diagnostic examinations (P = .92 and .40, respectively). Sensitivity was higher for masses than for nonmass enhancements (NMEs) (100% vs 85.3%; P = .009). Applying the conservative ADC cutoff achieved a sensitivity of 95.2% (60 of 63 lesions [95% CI: 86.7, 99.0]), with a biopsy reduction rate of 10.4% (25 of 240 lesions [95% CI: 6.7, 14.5]). Repeated single-reader measurements showed good agreement with clinical ADCs (intraclass correlation coefficient, 0.72 [95% CI: 0.58, 0.81]).

Conclusion

This study validated the clinical use of ADC cutoffs to reduce MRI-prompted biopsies by up to 16%, with a suggested tradeoff of lowered sensitivity for in situ and microinvasive disease manifesting as NME.

Clinical trial registration no. NCT02022579

© RSNA, 2024

Supplemental material is available for this article.

See also the editorial by Honda and Iima in this issue.

Summary

Apparent diffusion coefficient cutoffs can be feasibly and safely integrated into breast MRI interpretation to reduce unnecessary biopsies, and specific training and standardization may further improve diagnostic performance.

Key Results

• ■ In a retrospective evaluation of 240 Breast Imaging Reporting and Data System category 4 and 5 lesions at MRI with apparent diffusion coefficients (ADCs) collected at clinical interpretation and pathologic confirmation, applying the prespecified ADC cutoff (1.53 × 10⁻³ mm²/sec) would avoid 15.8% of biopsies while preserving high sensitivity (92.1% overall and 97.2% for invasive disease).

• ■ Performance for reducing biopsies was similar for lesions detected at high-risk screening and diagnostic examinations (16.7% vs 15.2%; P = .78).

• ■ Sensitivity was higher for masses than for nonmass enhancement lesions (100% vs 85.3%; P = .009).

Introduction

Breast cancer is the second leading cause of cancer-related death among women in the United States (1). Dynamic contrast-enhanced (DCE) MRI is recommended for supplemental screening of women at high risk of developing breast cancer (2–8). Emerging data suggest that supplemental MRI screening can increase early cancer detection in additional populations, especially intermediate-risk women with dense breast tissue (9,10). DCE MRI shows abnormal vascularity in developing breast cancers following the intravenous injection of a gadolinium-based contrast agent. This approach provides unrivaled sensitivity and cancer detection rates but moderate specificity, resulting in substantially increased false-positive rates and unnecessary biopsies compared with mammography (2,11). Associated patient anxiety and costs additionally limit patient acceptance of DCE MRI (12).

Diffusion-weighted imaging (DWI) is a complementary noncontrast MRI technique that has repeatedly demonstrated potential for improving diagnostic accuracy for breast cancer detection (13–15). DWI reflects the brownian motion (ie, diffusion) of water molecules in tissue. This enables characterization of lesion cellularity and microstructure, independent of contrast enhancement characteristics, that can be quantified using apparent diffusion coefficients (ADCs) (13,15). DWI is fast, taking only a few minutes; is widely available on various scanner platforms; and can be integrated into conventional breast MRI protocols to improve differentiation of benign and malignant breast lesions (14–16). Despite the growing body of evidence for its efficacy, clinical implementation of breast DWI has been limited to date, related to inconsistent standardization (in diffusion-weighted image acquisition and ADC generation) and lack of independent validation of ADC cutoffs in separate data sets (14,16).

Recently, the Eastern Cooperative Oncology Group–American College of Radiology Imaging Network, or ECOG-ACRIN, A6702 multicenter study (ClinicalTrials.gov identifier NCT02022579) identified an optimal diagnostic ADC cutoff using a standardized diffusion-weighted image acquisition protocol implemented across a range of scanner platforms and sites, centralized offline analysis by trained readers, and custom software tools. This cutoff was determined to potentially reduce overall biopsies by 20.9% (35.9% of biopsies with benign findings) without affecting sensitivity (14). However, because this cutoff was assessed within the same data set from which it was derived, it remains to be established whether this level of performance can be achieved when the cutoff is applied in the clinical setting with variable interpreting radiologists using commercially available picture archiving and communication system (PACS)–based software tools. The purpose of this study was therefore to validate the performance of the prespecified ADC cutoff determined by the multicenter A6702 trial for reducing unnecessary biopsies in lesions detected at clinical point-of-care breast MRI interpretations.

Materials and Methods

Patients

The institutional review board approved this retrospective study performed at a single academic medical center (Hutch institutional review no. 7339). The requirement for informed consent was waived for reviewing clinical images and medical records. Consecutive breast MRI examinations performed between May 2015 and January 2019 with a biopsy recommendation (Breast Imaging Reporting and Data System [BI-RADS] category 4 or 5 assessment [17]) followed by definitive biopsy outcome were included in this study. Clinical indications for MRI included high-risk screening characteristics (eg, increased family or genetic risk, personal history, previous high-risk pathologic finding) and diagnostic reasons (evaluation of extent of disease in women with recently diagnosed breast cancer, problem-solving for palpable lesion or other symptom, or short-term interval follow-up). Lesions were excluded for the following reasons: (a) lesions ipsilateral to a known cancer, (b) no pathologic confirmation, and (c) no recorded point-of-care ADC measurements. All clinical point-of-care breast MRI examinations were interpreted by one of several fellowship-trained breast radiologists (including J.M.L. and H.R., each with >10 years of breast imaging experience). Lesion ADCs were measured during clinical interpretations and recorded in the clinical database. However, these ADCs were not generally used for BI-RADS assessment because of a lack of consensus on appropriate cutoff to obviate biopsy. Lesion outcomes classified as benign or malignant based on pathology reports after breast biopsy or excision were used as the reference standard for diagnostic performance metrics. Age, race, ethnicity, and other factors were collected from electronic health records to describe the demographic characteristics of the study patients.

MRI Examinations

All breast MRI examinations were performed with a 3.0-T MRI scanner (Achieva TX; Philips Healthcare) with a dedicated 16-channel breast coil. Images were acquired in axial orientation, and each examination included T2-weighted, DWI, and DCE MRI sequences, in line with American College of Radiology breast MRI accreditation and European Society of Breast Imaging breast DWI guidelines (16,18) and following the A6702 protocol (full protocol [14] in Table S1). DWI scans were first spatially registered across b values to correct for patient motion and eddy current effects with use of onboard vendor software, and ADC maps were generated from images with b values of 0, 100, 600, and 800 sec/mm² by voxelwise linear regression fit of the classic monoexponential decay function.

Image Analysis and Clinical ADC Measurement

As part of standard clinical breast MRI interpretation at Fred Hutchinson Cancer Center since May 2015, radiologists were instructed to measure point-of-care ADCs for all lesions assessed as BI-RADS category 4 or 5, although this input was not forced, and the value could be left blank with an option for explanation. During the study period, radiologists performed this measurement using the PACS workstation (RA1000 Centricity; GE HealthCare) by manually placing a single two-dimensional region of interest (ROI) on the scanner-generated ADC map in the area corresponding to the enhancing lesion at DCE MRI (referencing the contrast-enhanced images). ROI size and shape were left to the clinician; however, the tools available on this PACS were limited to a round or oval ROI or single-pixel measurement. The mean ADC of the lesion ROI voxels was recorded from the PACS display into the clinical database along with BI-RADS characteristics, including lesion type (mass or nonmass enhancement [NME]) and maximum diameter (17). As described earlier, although the ADC was available, it was not typically used for final BI-RADS assessment and biopsy recommendation.

Reproducibility of ADC Measurements

To assess ADC measurement reproducibility, one radiologist (A.M.W., breast imaging fellow with 1 year of experience) who did not participate in the initial MRI interpretations independently remeasured the first 104 lesions identified at consecutive screening MRI examinations in the study sample, blinded to pathologic outcomes and previously recorded clinical ADC. These single-reader lesion ADCs were measured on a different PACS workstation platform (Universal Viewer, version 6.0; GE HealthCare) using an ellipse or freehand ROI tool (if necessary) and consistent approach generally targeting the lower ADC subregion identified by visual assessment.

Interpretation Thresholds

Two prespecified ADC cutoffs identified in the A6702 trial were tested: (a) an optimal cutoff for balancing sensitivity and biopsy reduction and (b) a conservative cutoff for prioritizing sensitivity. The optimal cutoff was 1.53 × 10⁻³ mm²/sec. The conservative cutoff was 1.683 × 10⁻³ mm²/sec (14), rounded in this study to 1.68 × 10⁻³ mm²/sec. For the primary analysis, the cutoffs were applied to all pathologically confirmed BI-RADS 4 or 5 lesions with clinical ADC measurements, with exploratory stratifications based on MRI indication, lesion type, and lesion size.

Statistical Analysis

All calculations were performed using R software, version 4.0.3 (R Foundation for Statistical Computing) by a biostatistician (D.S.H.). The threshold for statistical significance was set at two-sided P < .05. The nonparametric bootstrap was used to calculate 95% CIs and compare ADCs between benign and malignant lesions, ADCs between clinical and single-reader reads, and performance or reproducibility metrics between subgroups. Lesions were treated as clustered by individual rather than as fully independent observations, so bootstrap resampling was performed by individual to account for nonindependence of multiple lesions per individual (19). The Clopper-Pearson exact method for CIs was used when performance estimates were at or close to 100%. Clinical characteristics were compared by breast MRI indication (screening vs diagnostic) and whether clinical ADC was recorded or not using the Fisher exact test or the Wilcoxon rank-sum test (patient characteristics) or generalized estimating equation–based logistic regression (lesion characteristics) to account for the nonindependence of multiple lesions per patient.

Diagnostic performance of the ADC cutoffs was summarized using the following: (a) overall biopsy reduction rate, the proportion of all lesions that would not be biopsied after applying the ADC cutoff (ie, lesion ADC higher than the cutoff); (b) benign biopsy reduction rate, the proportion of benign lesions with ADC higher than the cutoff; and (c) sensitivity, the proportion of malignant lesions that would still be biopsied after applying the ADC cutoff (ie, lesion ADC less than or equal to the cutoff). The positive predictive value of biopsy recommendation using clinical breast MRI (PPV₂) was determined (17). Generalized estimating equation–based regression was used to compare sensitivity between masses and NMEs while adjusting for lesion size. The reproducibility of ADC measurements based on clinical point-of-care ADCs and repeated single-reader measurements was summarized using the intraclass correlation coefficient, within-patient SD (20), and Bland-Altman plots and metrics (mean difference and limits of agreement).

Results

Patient and Lesion Characteristics

During the study period, 299 BI-RADS 4 or 5 lesions detected at 255 breast MRI examinations in 247 women without ipsilateral known breast cancer underwent biopsy (Fig 1). Of the 299 pathology-confirmed lesions, 59 lesions (15 malignant) in 46 women were excluded because no point-of-care ADC was recorded. A reason for not measuring ADC was not given in most cases, although there was no evidence of differences in patient and lesion characteristics between examinations with and without ADC recorded (P = .19 to P > .99) (Tables S2 and S3, respectively). The final analysis data set included 240 BI-RADS category 4 or 5 lesions in 201 women (mean age, 49 years ± 13 [SD]; five with multiple examinations and 35 with multiple lesions) with clinical ADC measurements (median lesion size, 12 mm [IQR, 7–19 mm]). Patient and lesion characteristics are given in Tables 1 and 2, respectively. Clinical indications for breast MRI examinations were screening in 90 women (108 lesions) and diagnostic in 111 women (132 lesions). Pathologic assessment revealed 63 malignant lesions (PPV₂, 26.3% [63 of 240 lesions] [95% CI: 20.5, 32.0]); 26 were at screening examinations (PPV₂, 24.1% [26 of 108 lesions] [95% CI: 16.0, 32.2]) and 37 were at diagnostic examinations (PPV₂, 28.0% [37 of 132 lesions] [95% CI: 19.9, 36.2]). There were 24 invasive ductal carcinomas, 27 ductal carcinomas in situ, nine invasive lobular carcinomas, and three malignant neoplasms of other types (one malignant phyllodes and two metastatic renal carcinomas). There were 177 benign lesions (82 screening and 95 diagnostic), including 11 with high-risk abnormalities of atypical ductal hyperplasia, lobular carcinoma in situ, and atypical lobular hyperplasia (Table 2).

Figure 1:

Study inclusion flowchart. ADC = apparent diffusion coefficient, BI-RADS = Breast Imaging Reporting and Data System, DWI = diffusion-weighted imaging. * = Five women underwent multiple examinations, and 35 women had multiple lesions (201 women, 206 examinations, 240 lesions).

Table 1:

Patient Characteristics

Table 2:

Lesion Characteristics

Diagnostic Performance Applying Prespecified A6702 ADC Cutoffs

Clinical ADC measurements were higher for benign than malignant lesions (mean, 1.21 × 10⁻³ mm²/sec ± 0.38 vs 1.02 × 10⁻³ mm²/sec ± 0.32; P < .001) (Fig 2). Applying the optimal ADC cutoff of 1.53 × 10⁻³ mm²/sec resulted in an overall biopsy reduction rate of 15.8% (38 of 240 lesions [95% CI: 11.2, 20.9]) and benign biopsy reduction rate of 18.6% (33 of 177 lesions [95% CI: 12.8, 24.7]), with a sensitivity of 92.1% (58 of 63 lesions [95% CI: 82.4, 97.4]). Sensitivity for invasive cancers was 97.2% (35 of 36 lesions [95% CI: 82.7, 99.6]). Of 38 biopsies potentially averted with this cutoff, 33 lesions were benign and five were malignant (Table 3). The five false-negative findings included one microinvasive lobular carcinoma with extensive lobular carcinoma in situ and four ductal carcinomas in situ (two high-grade, one intermediate-grade, and one low-grade); two examples are shown in Figure 3. With use of the alternate conservative ADC cutoff of 1.68 × 10⁻³ mm²/sec, sensitivity was 95.2% (60 of 63 lesions [95% CI: 86.7, 99.0]), and the overall and benign biopsy reduction rates were 10.4% (25 of 240 lesions [95% CI: 6.7, 14.5]) and 12.4% (22 of 177 lesions [95% CI: 7.8, 17.7]), respectively (Table 3).

Figure 2:

Apparent diffusion coefficient (ADC) measurements for benign and malignant Breast Imaging Reporting and Data System (BI-RADS) category 4 or 5 lesions at MRI. (A, B) Benign BI-RADS category 4 lesion in a 55-year-old woman. (A) Dynamic contrast-enhanced (DCE) MRI scan shows a 12-mm, oval, circumscribed, homogeneous mass in the right posterior breast (arrow). (B) On the diffusion-weighted image, the lesion shows high diffusivity (arrow), with an ADC of 2.04 × 10⁻³ mm²/sec. Biopsy revealed benign fibroadenoma. (C, D) Malignant BI-RADS category 4 lesion in a 41-year-old woman. (C) DCE MRI scan shows a 24-mm irregular, heterogeneous mass at the 3 o’clock position in the left breast (arrow). (D) On the diffusion-weighted image, the lesion demonstrates low diffusivity (arrow), with an ADC of 0.91 × 10⁻³ mm²/sec. Biopsy revealed invasive ductal carcinoma. All MRI scans are shown in the axial plane. (E) Box plot shows ADC measurements for 240 BI-RADS 4 and 5 lesions (63 malignant and 177 benign). Both prespecified optimal and conservative ADC cutoffs are indicated on the plot for reference: 1.53 × 10⁻³ mm²/sec (dashed line) and 1.68 × 10⁻³ mm²/sec (dotted line), respectively. All ADCs were calculated with maximum b value of 800 sec/mm². The midline of each box shows the median value, and the box extends from first to third quartiles. The whiskers extend from the first and third quartiles to the smallest and largest values, respectively, if those values are within 1.5 times the IQR. If not, the length of the whiskers is limited to 1.5 times the IQR.

Table 3:

Diagnostic Performance of Clinical ADCs Using Predetermined A6702 Trial Cutoffs

Figure 3:

False-negative findings according to clinical apparent diffusion coefficient (ADC) (ADC greater than 1.53 × 10⁻³ mm²/sec). (A–C) Breast Imaging Reporting and Data System (BI-RADS) category 4 lesion detected at the 10 o’clock position in the right breast of a 47-year-old high-risk woman undergoing screening breast MRI. (A) Dynamic contrast-enhanced (DCE) MRI subtraction image shows a 70-mm segmental heterogeneous nonmass enhancement (NME) (oval). (B, C) On diffusion-weighted images (b = 800 sec/mm²), a small portion of the lesion shows high signal intensity (arrow, B), and the recorded clinical ADC was 1.57 × 10⁻³ mm²/sec (ellipse and arrow, C), just above the cutoff. MRI scans are shown in the axial plane. Excision revealed microinvasive lobular carcinoma with extensive noncancerous lobular carcinoma in situ. (D–F) BI-RADS category 5 lesion detected at the 6 o’clock position in the right breast of a 60-year-old woman undergoing MRI to evaluate extent of disease for newly diagnosed cancer of the left breast. (D) DCE MRI postcontrast image shows segmental NME with clustered ring internal enhancement measuring 58 mm (arrow). (E, F) On diffusion-weighted images (b = 800 sec/mm²), the lesion shows high signal intensity (arrow, E), and the recorded ADC was 1.79 × 10⁻³ mm²/sec (ellipse and arrow, F). MRI scans are shown in the axial plane. Final pathologic determination after surgical excision was intermediate-grade ductal carcinoma in situ. All ADCs were calculated with maximum b value of 800 sec/mm².

Stratified Performance of Clinical ADCs

After stratification by lesion type, sensitivity was higher for masses than for NMEs (100% [29 of 29 lesions] vs 85.3% [29 of 34 lesions]; P = .009), whereas overall (15.3% vs 16.4%; P = .80) and benign (20.0% vs 17.1%; P = .63) biopsy reduction rates were similar at the optimal ADC cutoff of 1.53 × 10⁻³ mm²/sec (Table 4). The higher sensitivity for masses versus NMEs persisted even after controlling for size (adjusted difference in sensitivity between masses and NMEs: 15.0% [95% CI: 2.4, 27.6]; P = .02) (Fig 4). Of note, all five ADC false-negative results were NMEs (three from screening examinations and two from diagnostic examinations).

Table 4:

Subgroup Analysis Using the Optimal ADC Cutoff of 1.53 × 10⁻³ mm²/sec as Determined by the A6702 Trial

Figure 4:

Subgroup analysis forest plots show mean estimates (dots) for performance metrics of sensitivity, overall biopsy reduction rate, and benign biopsy reduction rate, with 95% CIs indicated by horizontal bars. No evidence of performance differences was observed between screening and diagnostic examinations (P > .40 for all) or between lesions 10 mm or smaller and lesions larger than 10 mm (P > .14 for all). Sensitivity was higher for masses than for nonmass enhancement (NME) (P = .009), even after controlling for size (P = .02). The difference in sensitivity between masses and NME remained significant among lesions larger than 10 mm (100% [16 of 16] vs 86% [24 of 28]; P = .03) but not among lesions 10 mm or smaller (100% [13 of 13] vs 83% [five of six]; P = .74). ADC = apparent diffusion coefficient.

Other stratified analyses did not identify evidence of differences. Diagnostic performance of ADC was similar for screening and diagnostic examinations, including overall biopsy reduction rate (16.7% vs 15.2%; P = .78), benign biopsy reduction rate (18.3% vs 18.9%; P = .92), and sensitivity (88.5% vs 94.6%; P = .40). After stratification by lesion size, there was no evidence of a difference between lesions 10 mm or smaller and lesions larger than 10 mm in overall biopsy reduction rate (19.3% vs 12.7%; P = .14), benign biopsy reduction rate (22.1% vs 14.6%; P = .16), or sensitivity (94.7% vs 90.9%; P = .57). Performance was also explored according to BI-RADS category, although there were only 11 BI-RADS 5 lesions (nine of which were malignant). Sensitivity was similar for both BI-RADS 4 and 5 lesions (92.6% vs 88.9%; P = .86), with no evidence of a difference in the biopsy reduction rates (16.2% vs 9.1%; P = .45). Benign biopsy reduction rate could not be evaluated because there were only two benign BI-RADS 5 lesions. Subgroup results were similar when the alternate conservative ADC cutoff of 1.68 × 10⁻³ mm²/sec was used (Table S4).

Reproducibility of Lesion ADC Measurements

Lesion ADCs were remeasured by one reader for 104 consecutive lesions identified at screening examinations. There was good agreement between single-reader ADC and clinical point-of-care ADC, with an intraclass correlation coefficient of 0.72 (95% CI: 0.58, 0.81) and no evidence of a difference in mean ADC (mean, 1.19 × 10⁻³ mm²/sec ± 0.37 vs 1.19 × 10⁻³ mm²/sec ± 0.36; P = .97) (Table S5, Fig 5). The within-patient SD was 0.19 × 10⁻³ mm²/sec (95% CI: 0.15, 0.24). There was no evidence of a difference between masses and NME for the intraclass correlation coefficient (P = .33) and within-patient SD (P = .25) (Table S5). Diagnostic performance of single-reader ADC at the optimal ADC cutoff of 1.53 × 10⁻³ mm²/sec was similar to that of clinical ADC in the same group of lesions, with overall and benign biopsy reduction rates of 20.2% (21 of 104 lesions [95% CI: 12.3, 28.6]; P = .29 for single-reader ADC vs clinical ADC) and 24.4% (19 of 78 lesions [95% CI: 14.4, 34.6]; P = .13), respectively, and sensitivity of 92.3% (24 of 26 lesions [95% CI: 74.9, 99.1]; P = .77).

Figure 5:

Reproducibility of lesion apparent diffusion coefficient (ADC) measurements. (A) Dynamic contrast-enhanced (DCE) MRI scans in a 37-year-old woman undergoing screening breast MRI show a Breast Imaging Reporting and Data System category 4 mass (arrows). Clinical ADC was 1.19 × 10⁻³ mm²/sec ± 0.23 measured by the attending radiologist on the picture archiving and communication system workstation at the time of the examination, and retrospective single-reader ADC was 1.36 × 10⁻³ mm²/sec ± 0.20 (respective regions of interest are shown on the ADC map). Biopsy and excision revealed benign atypical ductal hyperplasia. (B) Bland-Altman plot shows comparison for clinical and single-reader ADC measurements from 104 consecutive screening examinations. ADC difference was calculated as clinical ADC minus single-reader ADC. The dotted line indicates the mean difference, the gray ribbon shows the 95% CI for the mean difference, and dashed lines represent the limits of agreement. Results are stratified by mass (yellow) and nonmass enhancement (NME) (pink) lesion type. There was good agreement between the two ADC measurements, with an intraclass correlation coefficient of 0.72 (95% CI: 0.58, 0.81) and no evidence of a difference in mean ADC (1.19 × 10⁻³ mm²/sec ± 0.36 for clinical ADC vs 1.19 × 10⁻³ mm²/sec ± 0.37 for single-reader ADC; P = .97). The within-patient SD was 0.19 × 10⁻³ mm²/sec (95% CI: 0.15, 0.24).

Discussion

This single-center retrospective study sought to validate the performance of point-of-care lesion apparent diffusion coefficient (ADC) measurements with predetermined ADC diagnostic cutoffs identified by the Eastern Cooperative Oncology Group–American College of Radiology Imaging Network Cancer Research Group multicenter A6702 trial (14) for reducing unnecessary biopsies at breast MRI. Our results confirmed that applying the prespecified optimal ADC cutoff of 1.53 × 10⁻³ mm²/sec or higher in clinical practice could reduce the benign biopsy rate by about 19% and overall biopsy rate by about 16% (similarly for screening and diagnostic examinations: 16.7% vs 15.2%, respectively; P = .78). High sensitivity was maintained (92% overall and 97% for invasive cancers), and the few missed cancers all manifested as nonmass enhancement with predominantly in situ disease. As expected, the higher, more conservative cutoff of 1.68 × 10⁻³ mm²/sec could better preserve sensitivity (95%), but with the tradeoff of fewer averted biopsies. Moreover, good interreader reproducibility of ADC measurements despite using different clinical picture archiving and communication system–based tools further supports the viability of eventual adoption and clinical implementation of diffusion-weighted imaging for breast MRI interpretation.

There have been numerous attempts to identify ADC cutoffs to differentiate benign and malignant lesions. However, the reported cutoffs in these studies varied (0.87–2.00 × 10⁻³ mm²/sec). The A6702 study established an optimal ADC cutoff from prospective multicenter and multivendor breast DWI data obtained from a prespecified study protocol (14). The trial identified an optimal ADC cutoff (1.53 × 10⁻³ mm²/sec, using a maximum b value of 800 sec/mm²) that could maximally reduce benign biopsies without lowering sensitivity. A subsequent retrospective study further supported the robustness of the A6702 findings, showing a potential 15.2% overall biopsy reduction rate and 96.6% sensitivity testing a similar ADC cutoff (1.50 × 10⁻³ mm²/sec) in a large heterogeneous multicenter data set (in terms of protocols and patient sample) (21). To date and our knowledge, no previous study has independently validated the A6702 cutoffs using point-of-care measurements collected in a real-world clinical practice setting.

Subgroup analyses by lesion type and size in our study identified a significantly higher sensitivity of the optimal cutoff of ADC of 1.53 × 10⁻³ mm²/sec or higher for masses than for NMEs. In general, NMEs can be less detectable or incorrectly characterized at DWI because of the lower spatial resolution and greater partial volume effects relative to T1- and T2-weighted sequences (15,21–23). Our findings confirm that this can be especially problematic for detecting ductal carcinoma in situ, which more often manifests as NME and made up the majority of DWI false-negative findings in our study.

Our study had several limitations. First, clinical ADCs were obtained during MRI interpretations and may have influenced the clinical BI-RADS assessments. This could have led to underestimation of biopsy reduction rates by skewing benign lesions in our study toward those exhibiting lower ADC. However, these data were collected before publication of the Eastern Cooperative Oncology Group–American College of Radiology Imaging Network A6702 trial findings, and ADC was recorded but rarely used for clinical decisions. Second, missing ADC measurements for 20% of MRI-detected BI-RADS category 4 or 5 lesions (59 of 299) may reflect low DWI evaluability, possibly overestimating real-world biopsy reduction rates. Third, the study was performed at a single institution with high breast MRI screening volumes, subspecialized breast imaging radiologists, and a standardized MRI protocol on a single scanner. Variability across scanner platforms and practice sites may affect performance at other institutions. Fourth, ROIs were inconsistently saved on the PACS, limiting our ability to retrospectively gain insights on ROI approach. Fifth, the pathology reference standard was not available for 17% of MRI-depicted BI-RADS category 4 or 5 lesions (60 of 359). Exclusion of these lesions may have introduced verification bias in diagnostic performance. Moreover, malignant findings did not include any mucinous carcinomas. This rare tumor type (1%–7% of breast cancers) exhibits higher ADC than other malignant lesions because of mucin contents and low cellularity (24) and could result in false-negative results based on an ADC cutoff.

In conclusion, our study results support adoption of apparent diffusion coefficient (ADC) cutoffs in the clinical setting, with the caveat that radiologists may need to be more cautious about using ADCs to avert biopsy of nonmass enhancement (NME) lesions. The study provides evidence that an ADC cutoff can be feasibly integrated into clinical breast MRI interpretations to decrease overall biopsies by up to 16%. Whereas the ADC cutoff appears especially robust for masses, some tradeoff in sensitivity was observed in the clinical setting for in situ and microinvasive disease manifesting as NME, particularly ductal carcinoma in situ, which may be acceptable if confirmed by future studies. More specific DWI training and greater dissemination of standardized ADC measurement methods and cutoffs may further improve DWI performance. In the future, larger studies should be conducted to investigate the rates of lesion DWI nonevaluability and performance in different malignant subtypes.