Prostate cancer (PCa) is the second most common cancer and fifth leading cause of cancer-related deaths in men worldwide, with an estimated nearly 1.4 million new cases and 375,000 deaths in 2020 (1). For men with high-risk localized and locally advanced PCa, multimodal therapy strategies, such as a reasonable first step of radical prostatectomy (RP) to reduce the tumor burden, followed by adjuvant androgen deprivation therapy or radiation therapy, are often necessary (2). Several studies have found considerable benefits for patients receiving adjuvant hormonal therapy (AHT) after RP compared to observations after RP (3–6). However, the current definition of high-risk prostate cancer includes a heterogeneous group of patients with different outcomes based on different baseline characteristics (7). A proportion of patients may exhibit early resistance to AHT and elevated PSA levels in the short term after RP, which has given rise to studies of prognostic factors, including novel biomarkers such as advanced magnetic resonance imaging (MRI) parameters.

In recent years, MRI has been widely used in clinical practice for the detection and staging of PCa (8) and is also promising in predicting the prognosis of PCa (9, 10). The apparent diffusion coefficient (ADC), derived from diffusion-weighted MRI, is thought to be related to the cellularity and interstitial structure of pathological tissue and therefore reflects the histopathological heterogeneity of malignant tumors (11–13). Histogram analysis is a texture-based statistical measurement method for images that can quantitatively provide a series of information on ADC maps, such as percentile ADC values, kurtosis, and skewness. Numerous studies have shown that ADC histogram analysis derived from baseline MRI examination has prognostic value in several types of tumors, including glioblastoma (14), esophageal carcinoma (15), and squamous cell cervical cancer (16). In PCa, ADC histogram analysis has recently been declared to be valuable in the detection, differentiation, and aggressiveness assessment of cancer lesions. Peng et al. (17) found that ADC₁₀ and ADC_mean could differentiate between prostate cancer and normal tissue. Donati et al. (18) evaluated the correlation between the 10^th percentile, 25^th percentile, mean, and median ADC values and Gleason scores of 6, 7 and 8 and found the strongest correlation between the 10th percentile ADC values and Gleason score. Given that histogram parameters may reflect tumor heterogeneity, it is of clinical relevance to know whether histogram analysis can serve as a biomarker for the prediction of BCR and whether ADC parameters have an incremental prognostic value in patients with PCa receiving RP and AHT.

To our knowledge, the predictive effect of ADC histogram analysis on the prognosis of high-risk PCa patients receiving AHT after RP remains unclear. Therefore, the purpose of this study was to evaluate the prognostic value of preoperative ADC histogram analysis in this certain population and to develop a prognostic model combining clinical factors and ADC histogram parameters.

A total of 62 patients (median age: 66.50 years and IQR: 61.75-71.00 years) with pathologically confirmed prostate cancer were included. All patients were followed up for more than 3 years with a median follow-up period of 54.1 months (range, 41.1–79.3 months). BCR was found in 19 patients (30.6%). The BCR-fs rates at 12, 24 and 36 months were 85.5%, 79.0%, and 71.0%, respectively. The baseline characteristics of the enrolled patients are shown in Table 1 .

ADC histogram parameters demonstrated good to excellent reproducibility (ADC₁₀, ICC: 0.892; ADC₉₀, ICC: 0.917; ADC_mean, ICC: 0.803; ADC₅₀, ICC: 0.931; kurtosis, ICC: 0.867; skewness, ICC: 0.820).

In the univariate survival analysis, lower ADC values (ADC_mean ≤ 8.59 × 10⁻⁴ mm²/s, ADC₁₀ ≤ 6.18 × 10⁻⁴ mm²/s, ADC₅₀ ≤ 8.28 × 10⁻⁴ mm²/s, ADC₉₀ ≤ 11.7 × 10⁻⁴ mm²/s) and higher kurtosis (> 9.78×10⁻¹) were significant predictors of poorer BCR-fs (all p<.05, Table 2 ). For clinical parameters, SM R1, TPSA ≥ 24.6 ng/ml, and pT3~pT4 had significant relationships with poor BCR-fs (p = 0.011, p = 0.002, and p = 0.037, respectively). BCR-fs showed no significant differences in age, Gleason score, and pN ( Table 2 ). Kaplan−Meier curves of pT, SM, TPSA ADC₁₀, ADC₅₀, and ADC_kurtosis for BCR-fs are shown in Figure 2 .

Multivariate Cox regression analysis was performed for ADC_mean, ADC₁₀, ADC₅₀, ADC₉₀, and kurtosis histogram parameters, as well as preoperative TPSA, pT and SM status. ADC₅₀ and kurtosis remained significant (HR: 0.172, 95% CI: 0.055-0.541, p = 0.003; HR: 7.058, 95% CI: 2.288-21.773, p = 0.001;, respectively) after adjustment for clinical factors. The clinical parameters of preoperative TPSA and SM were independent predictors of BCR-fs (HR = 3.824, 95% CI = 1.406-10.402, p = 0.009; HR = 3.067, 95% CI = 1.110-8.477, p = 0.031, respectively).

Then, the clinical model was established with TPSA and surgical margins, while the clinical imaging model was established with ADC₅₀ and kurtosis as covariables. All variables in the clinical model remained significant predictors in multivariate prognostic models of BCR-fs (p<.100 for all), as shown in Table 3 . In the comparison of model performance, the C index of the clinical-imaging model was higher than that of the clinical model (C index = 0.85, 95% CI: 0.76-0.94; C index = 0.73, 95% CI: 0.61-0.84, respectively, p = 0.032). The inclusion of histogram parameters improved the diagnostic accuracy of the models for 12-, 24- and 36-month BCR-fs (AUC of clinical imaging model vs clinical model: 0.869 VS 0.733, p = 0.19; 0.859 VS 0.678, p = 0.038; and 0.921 VS 0.7557, p = 0.014; for 12-month, 24-month and 36-month BCR-fs, respectively), in which the improvement in the diagnostic accuracy of BCR-fs at 24 and 36 months was statistically significant ( Figure 3 ).

In this study, we evaluated the prognostic value of the ADC histogram derived from preoperative DWI in predicting the BCR of patients with high-risk PCa who underwent RP followed by AHT. The results showed that the ADC₅₀ and kurtosis of the ADC histogram were independent prognostic factors for BCR-fs, and when ADC parameters were incorporated into the clinical model, the prediction accuracy for BCR-fs was improved.

Several tools have been proposed to identify patients at risk for BCR, including preoperative tools, such as the Kattan Nomogram (22) and Prostate Cancer Risk Assessment (CAPRA) score (23), and postoperative tools, such as the Postsurgery Prostate Cancer Risk Assessment (CAPRA-S) score (24, 25). These assessments showed that factors including PSA and SM status were significantly associated with the risk of BCR (24). In our study, we also found that preoperative TPSA, pT stage and SM status were associated with the risk of BCR. Although most studies have demonstrated that the Gleason score is a risk factor for BCR, our study did not find a significant association between the Gleason score and the risk of BCR, similar to the results from a prospective, multicenter cohort of patients with high-risk localized and locally advanced prostate cancer in China who also received adjuvant hormone therapy after radical prostatectomy (26). Moreover, the BCR rate at the 24-month follow-up in our study was 21%, which is also similar to the BCR rate (17.4% at the 24-month follow-up) in that prospective study (26). We did not find any relationship between pN stage and BCR. The differences in the results between our study and other studies may be because of the heterogeneity of the study population and the limited sample size and follow-up time.

MRI is an essential clinical examination for prostate cancer. The diffusion-weighted imaging of MRI is a functional sequence that reflects the degree of restricted diffusion of water and is thought to be related to the cellularity and interstitial structure of pathological tissue. Tumors with different biological behaviors, such as metastasis and recurrence potential, may show different degrees of diffusion restriction. Attempts have been made to integrate MRI into prognostic models to improve the predictive performance (9, 10). Haralick features on T2WI have been previously shown to be associated with BCR in PCa (27, 28). However, the information embodied in the T1 and T2 sequences is more anatomical than functional, and not all patients underwent dynamic contrast-enhanced examination, so we chose DWI as the research object, similar to other histogram analysis studies of different cancers. Bourbonne et al. (29) used radiomics to analyze BCR and BCR-free survival in 107 high-risk PCa patients (pT3–pT4, positive margins, GS 8–10) treated with RP. They found that a texture feature, small zone emphasis (SZE_GLSZM) from the ADC map, was strongly correlated with the risk of BCR, independent of other radiomic features and clinical variables. However, the study reported that the MRI was performed on two different MRI scanners: a Phillips 3T and a Siemens 1.5T, which could have some relevant impact on the results. Compared to high-order texture features, which may involve complex statistical computations and less interpretability, first-order textural analysis can be better understood and may have more clinical validity (30).

ADC histogram analysis can yield additional parameters, such as different percentiles of ADC values, skewness, and kurtosis, which can provide more diffusion information than qualitative evaluation and can better reflect tumor heterogeneity. In this study, we evaluated percentile values of ADC₁₀ and ADC₉₀, because they were far from the ADC_mean, may provide information that the ADC_mean cannot, and are less susceptible to extreme values, like ADC_max and ADC_min. Rosenkrantz et al. (31) found that ADC_0-10, the mean of the bottom 10th percentile ADC, rather than the average value of the ADC, was a significant independent predictor for BCR in the multivariate logistic regression analysis. Donati et al. (18) also observed that the AUC of the 10^th percentile ADC was significantly higher than the mean ADC to predict the Gleason score. They argued that the 10^th percentile ADC was more representative of focal areas of high cellularity than the mean or median values, which were more susceptible to normal glandular tissue (18). However, according to our results, ADC₅₀ remained a significant parameter after stepwise backward regression to rule out collinearity of other ADC percentile values, which means that ADC₅₀ predicts BCR-fs better than ADC₁₀ in PCa patients after RP and AHT. While the reason remains unclear, one possible explanation for the difference between the results of this study and the previous ones is that we studied a different population. We enrolled patients with high-risk localized or locally advanced PCa whose cancer foci may be more prone to necrosis and hemorrhage. When delineating the ROI, micronecrosis and microhemorrhage lesions within the tumor were inevitable, leading to increased ADC values. We assume that micronecrosis and microhemorrhage also play a role in the prognostic value of the ADC histogram; thus, higher percentile ADCs may more comprehensively predict the outcome than lower percentile ADCs. These controversial results indicated the need for further investigation of the mechanisms underlying the association between different percentile ADCs and prognostic values.

Kurtosis is an important indicator of tumor homogeneity and has been reported to be associated with the prognosis of several cancers. Higher kurtosis on DWI was associated with poorer overall survival and relapse-free survival in surgically treated colorectal cancer patients (32). In thyroid cancer, ADC kurtosis was associated with the aggressiveness of the lesion, and higher kurtosis predicted lymphatic metastasis (33). In addition, kurtosis of the ADC histogram could serve as a potential biomarker for the prediction of tumor aggressiveness and survival in esophageal cancer (15). We speculate that the profound changes in the microarchitecture of distinct tumor cell clusters may be related to their recurrence potential, which is reflected in the corresponding changes in the ADC histogram.

In this study, we hypothesized that the prognostic value might increase when ADC histogram parameters were added to traditional clinical factors. Further comparison of the performance of the clinical imaging model and clinical model showed that the C index increased and the p value was 0.032. We also observed an increase in the AUC of the diagnostic accuracy of BCR-fs at 24 months and 36 months. This trend leads us to believe that ADC histogram analysis has incremental prognostic value in the prognosis of PCa patients.

This study has several limitations. First, this single-center, retrospective study could have some possible bias, which needs to be validated in multicenter prospective studies to provide more reliable evidence for clinical application. Second, the sample size is small, with only 62 patients’ survival information available, and it is difficult to set aside an independent validation cohort, which may not have sufficient power to draw conclusions. An independent cohort to validate this finding is required in the future. Studies with larger sample sizes may produce more accurate predictions. In addition, the follow-up time was relatively short, with longer follow-up times required to predict 5-year and 10-year BCR-fs, progression-free survival and overall survival. Besides, the delineation of the tumor ROIs was subjectively free-hand, which may reduce the research reproducibility. A standardization and consensus of the DWI protocol, including the imaging postprocessing method, especially automatic or semiautomatic segmentation, are warranted. Last but not least, this prostate prognostic model was constructed based on high-risk localized and locally advanced PCa, the community-based benefits of this model are limited.

Preoperative MRI-derived ADC histogram analysis has incremental prognostic value as an imaging biomarker for the prognosis of patients with high-risk PCa receiving AHT after RP. Our study results demonstrate that ADC₅₀ and kurtosis are the most valuable histogram-derived parameters for predicting BCR-fs in high-risk PCa patients.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the Tongji Hospital of Huazhong University of Science and Technology (Wuhan, China) ethics review committee. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

XM and HL had full access to all the data used in this study and takes responsibility for the accuracy of the data analysis. KH and YZ contributed equally to this work. Study concept and design, XM and HL. Drafting of the manuscript, KH and YZ. Critical revision of the manuscript for important intellectual content, all authors. Image obtaining and postprocessing, XM and PL. Statistical analysis, SL, GY, and QZ. Administrative, technical, or material support, HL and ZL. Study supervision, ZL. All authors contributed to the article and approved the submitted version.