This study aimed to assess the clinical feasibility of employing synthetic diffusion-weighted (DW) images with different b values (50, 400, 800 s/mm²) for the prostate cancer patients with the help of three models, namely CycleGAN, Pix2PiX, and DC²Anet. DW images of 170 prostate cancer patients were used to train and test models. Here, 119 patients were assigned to the training set and 51 patients to the testing set according to a ratio of 7:3. To generate synthetic b value DW images based on CycleGAN, Pix2Pix, and DC²Anet networks, three experiments were performed as follows: generating synthetic DW images with b values of 400 and 800 s/mm² from acquired DW images with b value of 50 s/mm²; generating synthetic DW images with b value of 800 8 s/mm² from acquired DW images with b value of 400 s/mm². Five metrics were used to compare real and synthetic b values. These metrics included Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Pearson’s Correlation Coefficient (PCC), Peak-Signal-to-Noise-Ratio (PSNR), and Structural Similarity Index Measure (SSIM). As well as, ADC values for different b values were computed using the mono-exponentially mode. The whole prostate volume was manually segmented by drawing regions of interest (ROIs) in each slice of the ADC maps. P values less than 0.05 were considered statistically significant. Based on the quantitative evaluation and for all metrics, especially for generating b values of 400 and 800 s/mm² from a b value of 50 s/mm², the DC²Anet model was found accurate and it outperformed CycleGAN and Pix2Pix models (P < 0.05). It is necessary to mention that the agreement between synthetic ADC (sADC) and real ADC (rADC) was satisfactory. No significant difference was observed in the one-way ANOVA between sADC and rADC in the whole prostate volume (P > 0.05). Our results showed the significant potential of the three used models for generating images with different b values in the case of prostate cancer. The results demonstrated that the used three models were accurate and robust for generating DW images and also, they outperformed other methods mentioned in the literature review.

本研究旨在评估在三种模型^（即CycleGAN 、Pix2PiX 、和 DC ² Anet。170 名前列腺癌患者的 DW 图像用于训练和测试模型。在这里，按照 7:3 的比例将 119 名患者分配到训练集，将 51 名患者分配到测试集。为了生成基于 CycleGAN、Pix2Pix 和 DC ^{2 Anet 网络的合成}b值 DW 图像，进行了如下三个实验： 生成b值分别为 400 和 800 s/mm ^{2的合成 DW 图像}从获得的b值为 50 s/mm ²的 DW 图像中获取；从获得的b值为 400 s/mm ^{2的 DW 图像生成}b值为 800 8 s/mm ²的合成 DW 图像。五个指标用于比较真实和合成b值。这些指标包括平均绝对误差 (MAE)、均方根误差 (RMSE)、皮尔逊相关系数 (PCC)、峰值信噪比 (PSNR) 和结构相似性指数测量 (SSIM)。此外，不同b的 ADC 值使用单指数模式计算值。通过在 ADC 图的每个切片中绘制感兴趣区域 (ROI) 来手动分割整个前列腺体积。P值小于0.05被认为具有统计学意义。基于定量评估和所有指标，特别是从50 s/mm ^2的b值生成 400 和 800 s/mm ²的b值，发现DC ² Anet 模型是准确的，并且优于 CycleGAN 和 Pix2Pix 模型（磷 < 0.05)。值得一提的是，合成ADC（sADC）和真实ADC（rADC）的一致性令人满意。sADC和rADC在整个前列腺体积中的单因素方差分析没有显着差异（P  > 0.05）。我们的结果表明，在前列腺癌的情况下，三种使用的模型在生成具有不同b值的图像方面具有巨大的潜力。结果表明，所使用的三个模型在生成 DW 图像方面是准确且稳健的，并且它们优于文献综述中提到的其他方法。