Uterine myomas can cause infertility. Studies are attempting to determine the indications for myomectomy. However, the multiplicity and localization of myomas complicate this issue. We aimed to develop a visualization tool to aid patients with infertility in their decision-making for myomectomy. We included 191 women with uterine myoma attending an outpatient infertility clinic, of whom 124 patients underwent myomectomy. Of these, 65 (52.4%) patients became pregnant within 17.6 months after surgery, and 54 (83.1%) of them had a live birth. A logistic regression model predicting the pregnancy rate (area under the curve, 0.82; 95% confidence interval, 0.74–0.89; validation value, 74.6%) was generated using the leave-one-out cross-validation method. This model incorporated five factors: age, maximum level of infertility intervention following myomectomy, presence of submucosal myoma, maximum diameter of the myoma, and type of myomas (multiple or single). We successfully visualized the degree of involvement of each factor in the pregnancy rate by developing a nomogram based on this model. We expanded the data from the preoperative magnetic resonance images and applied machine learning using a convolutional neural network. The classification accuracy was 71.4% for sensitivity and 77.7% for specificity. Heatmap images, generated using gradient-weighted class activation mapping to show the classification results of this model, could distinguish between myomas that required enucleation and those that did not. Although a larger sample size is needed to further validate our findings, this innovative pilot study demonstrates the potential of machine learning to refine assessment criteria and improve patient decision-making.

子宫肌瘤可导致不孕。研究正在尝试确定子宫肌瘤切除术的适应症。然而，肌瘤的多样性和局限性使这个问题变得复杂。我们的目标是开发一种可视化工具来帮助不孕患者做出子宫肌瘤切除术的决策。我们纳入了 191 名在不孕不育门诊就诊的患有子宫肌瘤的女性，其中 124 名患者接受了子宫肌瘤切除术。其中，65例（52.4%）患者在术后17.6个月内怀孕，其中54例（83.1%）活产。使用留一交叉验证方法生成预测妊娠率的逻辑回归模型（曲线下面积，0.82；95% 置信区间，0.74-0.89；验证值，74.6%）。该模型纳入了五个因素：年龄、肌瘤切除术后不孕干预的最大水平、粘膜下肌瘤的存在、肌瘤的最大直径和肌瘤的类型（多发或单发）。通过基于该模型开发列线图，我们成功地可视化了每个因素对妊娠率的参与程度。我们扩展了术前磁共振图像的数据，并使用卷积神经网络应用了机器学习。分类准确度的敏感性为 71.4%，特异性为 77.7%。使用梯度加权类激活映射生成的热图图像可以显示该模型的分类结果，可以区分需要摘除的肌瘤和不需要摘除的肌瘤。尽管需要更大的样本量来进一步验证我们的研究结果，但这项创新的试点研究证明了机器学习在完善评估标准和改善患者决策方面的潜力。