Diffusion-weighted imaging in monitoring the pathological response to neoadjuvant chemotherapy in patients with breast cancer: a meta-analysis

Neoadjuvant chemotherapy (NACT), since its first appearance until nowadays, has become a standard therapy for patients with breast cancer. It is suggested to have beneficial effect especially on locally advanced or inflammatory breast cancer [1]. The major benefit of NACT is to reduce the tumor size and to downstage the tumor burden, which may lead to the successful performance of breast-conserving surgery instead of mastectomy [2]. In addition, assessing the treatment responses to NACT can also help to determine the right time to perform the operation or to adjust the therapy regimen in case of an unfavorable tumor response at an early stage [3]. It is well-established in some previous studies that the response to NACT is correlated with long-term outcomes for breast cancer patients. Studies also reveal that pathological complete response (pCR) patients may have a superior chance to achieved disease-free survival and overall survival [4‐6]. Nevertheless, only a minority of patients were featured with pCR due to the heterogeneity of breast cancer. We could not accurately observe the pCR until the definitive breast surgery was completed, which always led to inappropriate surgery decision-making for patients [7, 8]. Therefore, it is crucial to find an effective method to separate the patients who have achieved pCR from pathological non-responders (pNR) before surgery.

Mammography, ultrasonography, positron emission tomography-computed tomography (PET/CT) and magnetic resonance imaging (MRI) are the most commonly applied conventional imaging techniques for the detection of NACT responses. Previous studies found that MRI was superior to mammography or ultrasonography in evaluating therapeutic response of NACT in breast cancer [9, 10]. A meta-analysis demonstrated a higher sensitivity in PET/CT and a higher specificity in MRI for the assessment of pCR [11]. Currently, contrast-enhanced magnetic resonance imaging (DCE-MRI) is frequently and commonly used for tumor response evaluation after NACT. However, the information provided by DCE-MRI regarding blood flow and vessel permeability might cause difficulty in differentiating viable residual cancer form surrounding scar, necrosis, fibrosis, or reactive inflammation resulting from NACT response. Thus, DCE-MRI has deficiencies for the examination of pathological response to NACT [12].

Diffusion-weighted imaging (DWI), with its unique tissue contrast mechanism, is regarded as a potential modality to overcome the limitations of traditional DCE-MRI evaluation [13]. DWI reveals the thermally driven motion of water molecules in the target tissue. It offers information concerning the integrity of cell membranes and cancer cellularity. The apparent diffusion coefficient (ADC), which can be quantified and measured on DWI, represents the complex diffusion of water in tissues [14]. With those characteristics, DWI can be sensitive in detecting the changes in the intratumor induced by NACT [15].

The accuracy of conventional imaging modality including MRI, PET/CT, mammography, and ultrasonography in the assessment of the pCR to NACT has been investigated by several recent meta-analyses [16‐19]. However, no previous study has focused on analyzing the performance of DWI in detecting the pCR in breast cancer to NACT systematically. In researches providing DWI evaluation, the data were limited. They only involved a small amount of studies which might weaken the statistic power of the analysis. By combining all available data, the present meta-analysis intended to evaluate the diagnostic role of DWI in monitoring pCR in breast cancer to NACT.

DWI, with its rapid, non-invasive, and without the use of contrast agent characteristic, has emerged as a practical mean to overcome the limitation of DCE-MRI [23, 39]. However, to our knowledge, no previous meta-analysis focused on evaluating the diagnostic performance of DWI in detecting patients’ complete response to NACT in breast cancer. Thus, we designed the current meta-analysis specific for this purpose. By combining data from 20 studies, we detected a 0.89 sensitivity and a 0.72 specificity for DWI, which indicated that DWI could be a valuable imaging method for assessing pCR to NACT in breast cancer. An approximately 0.91 AUC value, which was close to 1 (the perfect test result), also indicated an ideal diagnostic performance. The DOR of a test can serve as a single summary measure since it obtains from combining different sensitivity and specificity. It is defined as the ratio of the odds of positivity in disease relative subjects to the odds of positivity in the non-diseased [40]. The DOR value display in a wide range from 0 to infinity. A higher DOR value represents a better ability for the discrimination of the test performance. The outcome of our study has shown that the DOR estimated for DWI was 27.00 (95% CI, 15.60–46.73). This benign high-DOR value indicated that DWI could monitor pCR in NACT accurately.

Although the pooled statistic of our study implied that DWI might accurately detect pCR for breast cancer to NACT, significant heterogeneity in sensitivity (I² = 62.7%) and specificity (I² = 84.9%) were also noticed. The Spearman correlation coefficients of DWI (0.565, P > 0.05) already eliminated the threshold effect had on DWI evaluation. However, the considerable heterogeneity might be attributed to many other factors, such as variations in definition to separate responders from non-responders, variations in the duration of the study population, or differences in pathologic complete response rate or study designs. To reduce the influence induced by these diversities, we carried out several subgroups analyses concerning pCR rate, time duration of study population, and study design. Meta-regression analyses revealed no significant difference among the three subgroups. This finding implied that, although heterogeneity might exist between different studies, results across studies were still comparable with little or no differences outcomes.

Abundant studies have been conducted to evaluate the efficiency of DCE-MRI in diagnosing pathologic response to NACT for patients with breast cancer. Yet, only a few has investigated the DWI diagnostic accuracy for predicting pCR to NACT. Gu et al. [11] suggested that the sensitivity and specificity of DWI were 93 and 85%, respectively. Another meta-analysis, Wu et al. [41] reported a 93% sensitivity and a 82% specificity for DWI evaluation. It seems that both studies have a slightly higher sensitivity and a much higher specificity than our research. However, the two previous analyses only included a small amount of studies which provided DWI data. Gu et al. enrolled eight studies, and Wu et al. had six. Our study analyzed up to 20 groups of DWI data, and this might be the reason causing discrepancy in our result. Interestingly, we can observe in all three studies that the sensitivity of DWI is higher than the specificity. This finding might further prove the hypothesis that DWI could accurately assess pCR in sensitivity. However, it might lack specificity.

Generally, by predicting the outcome and identifying the pCR to NACT treatment, breast cancer patients can avoid inappropriate chemotherapy at early stage as well as additional toxic therapies, and hold a better chance to achieve pCR [42, 43]. Therefore, some researchers argue that it is crucial to find a specific time for DWI evaluation. However, the previous studies and our analysis all failed to find an exact time to perform DWI. The timing of DWI assessment in our research was varied from study to study. Many studies conducted DWI at several time points. Five studies performed DWI after 1 cycle of therapy, three conducted after 2 cycles, and four studies assessed after 3 cycles. The available data were limited and hampered us to perform subgroup analysis. Nevertheless, a pattern can still be observed from the included studies. It seems that the first 3 cycles might be the preferable timing for DWI assessment. This hypothesis needs further approval.

Some limitations should be taken into account in our analysis. First, a majority of the included studies contained a relatively small patients’ population which might weaken the statistical power of the study and might bring about inconclusive and imprecise results. Although quality assessment and publication bias test confirmed that the included studies were eligible, the effect brought by different sample size still could not be neglected. Second, patients with different breast cancer subtypes would be assigned with different treatment regimens which might eventually lead to different pathological responses [24, 25, 33]. Bufi et al. remarked that DWI might achieve a better diagnostic performance in luminal and hybrid tumor subtypes [24]. Another study by Bufi et al. reported that pretreatment ADC was capable of detecting pCR in Triple negative and HER2⁺ tumors [25]. Richard et al. found that luminal A and B subtypes had a lower pretreatment ADC than triple-negative tumors which indicated a superior performance of DWI in the prediction of pCR to NACT in triple-negative tumors [33]. Thus, subgroup analysis based on tumor phenotypes is desirable. However, the limited information, which only three studies had provided data of breast cancer subtypes, prevented us to conduct subgroup analysis on this aspect. Moreover, the definition of pCR could be a reason affecting the diagnostic accuracy test. Since too many various pCR definitions were applied in the included studies, such as Miller-Payne grading system, Mandard’s TRG criteria, Japanese Breast Cancer Society criteria, Chevalier-Sataloff classifications and classification by user-defined, subgroup comparison could not be performed in our study. Yet, although subgroups and threshold effect evaluation has diminished the influence of heterogeneity, the effect of heterogeneity still cannot be eliminated completely. Several variables, regarding treatment regiments of NACT, timing of pCR evaluation, standards and pattern of DWI measurement, and the optimal cut-off values of diagnosis, should be taken into consideration. However, the information retrieved from the included studies were limited and inconsistent regarding the above factors, making it impossible to conduct subgroup analyses to eliminate their effect.

Despite some limitations, the findings of our study indicate that DWI modality holds a relatively high sensitivity and specificity for the evaluation and prediction of pCR of breast cancer to NACT. The result of our analysis suggests that the application of DWI in combination with other imaging modality may yield greater precision and accuracy in assessing the pCR after NACT.