Investigation on clinical risk factors of bladder lesion by machine learning based interpretable model

Yunxin Wang, Jun Li, Yunfeng Song, Hui Wei, Zejun Yan, Shuo Chen, Zhe ZhangScientific 2024 · DOI

Abstract

Bladder lesion commonly occurs in patients with benign prostatic hyperplasia (BPH), and routine screening is vital for timely detection. Current clinical methods like cystoscopy are invasive and inconvenient for routine use. This study proposes a non-invasive screening method using machine learning to diagnose bladder lesions based on routine clinical risk factors. Using a cohort of 253 BPH patients, the model identified bladder compliance, maximum flow rate (Qmax), prostate-specific antigen (PSA), and postvoid residual (PVR) as the primary predictors. The Support Vector Machine (SVM) model achieved high precision (86.83%) and interpretability via SHAP analysis, offering a potential routine screening tool.

Introduction

Bladder lesions, including diverticula and trabeculation, are frequent complications of Benign Prostatic Hyperplasia (BPH). If undetected, these can progress to severe conditions like acute renal failure or bladder cancer. The current diagnostic "gold standard," cystoscopy, is invasive and uncomfortable, making it unsuitable for frequent screening. While imaging like MRI or CT exists, costs and technical requirements limit their routine application. This research addresses the need for a low-cost, non-invasive "white-box" screening tool using standard clinical metrics.

Methods

The study utilized a retrospective cohort of 253 male BPH patients (172 with lesions, 81 controls). Six routine clinical factors were analyzed: bladder compliance, Qmax, PSA, PVR, detrusor contractility, and IPSS/QoL scores.

The AI-Assisted Diagnosis workflow, from clinical feature collection to model interpretability analysis.

The researchers compared six machine learning architectures, including Support Vector Machines (SVM), Random Forest, and Multi-Layer Perceptron. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. Crucially, the team integrated SHAP (Shapley Additive Explanations) to provide interpretability, allowing clinicians to understand the specific factors driving each diagnostic prediction.

Results

The SVM model emerged as the top performer, balancing high accuracy with clinical reliability. It achieved a precision of 86.83% and an Area Under the Curve (AUC-ROC) of 0.8055.

Performance metrics of the diagnostic model, including the ROC curve and key evaluation indices like F1, Recall, and Precision.

Beyond raw performance, the study identified the "Big Four" predictors: bladder compliance, Qmax, PSA, and PVR. Threshold analysis pinpointed specific clinical "tipping points" for increased risk, such as a PVR greater than 172.49 cc or a Qmax below 8.13 ml/s. Subjective symptom scores (IPSS) were found to be the least reliable predictors.

Discussion and Future Work

While promising, the study's single-center nature and relatively small sample size necessitate further validation in multi-center trials. Future work aims to incorporate urinary biomarkers to enhance accuracy and explore deep learning models to capture more complex clinical relationships. This interpretable framework represents a significant step toward non-invasive, routine screening for bladder lesions in BPH patients.

Summary updated

Abstract

Bladder lesion commonly occurs in patients with benign prostatic hyperplasia (BPH), and the routine screening of bladder lesion is vital for its timely detection and treatment, in which the risk of bladder lesion progression can be effectively alleviated. However, current clinical methods are inconvenient for routine screening. In this study, we proposed a convenient routine screening method to diagnose bladder lesions based on several clinical risk factors, which can be obtained through non-invasive, easy-to-operate, and low-cost examinations. The contribution of each clinical risk factor was further quantitatively analyzed to understand their impact on diagnostic decision-making. Based on a cohort study of 253 BPH patients with or without bladder lesions, the proposed diagnostic model achieved high accuracy using these clinical risk factors. Bladder compliance, maximum flow rate (Qmax), prostate specific antigen (PSA), and postvoid residual (PVR) were identified as the four most important clinical risk factors. To the best of our knowledge, this is the innovative research to predict bladder lesions based on the risk factors and quantitatively reveal their contributions to diagnostic decision-making. The proposed model has the potential to serve as an effective routine screening tool for bladder lesions in BPH patients, enabling early intervention to prevent lesion progression and improve the quality of life.
BibTeX
@inproceedings{wang2024investigation,
  author = {Yunxin Wang and Jun Li and Yunfeng Song and Hui Wei and Zejun Yan and Shuo Chen and Zhe Zhang},
  title = {{Investigation on clinical risk factors of bladder lesion by machine learning based interpretable model}},
  booktitle = {Scientific},
  year = {2024},
  doi = {10.1038/s41598-024-75104-x},
}
Figure