Deep Learning-Based Uroflowmetry Curve Analysis Improves the Noninvasive Diagnosis of Lower Urinary Tract Symptoms
Purpose: This study aimed to evaluate the performance of an artificial intelligence (AI)-based analysis of uroflowmetry (UFM) curve images, enhanced with customized preprocessing techniques, to improve diagnostic accuracy for bladder outlet obstruction (BOO) and detrusor underactivity (DUA). Methods...
Gespeichert in:
| Veröffentlicht in: | International neurourology journal S. 73 - 82 |
|---|---|
| Hauptverfasser: | , , , , , , |
| Format: | Journal Article |
| Sprache: | Englisch |
| Veröffentlicht: |
대한배뇨장애요실금학회
01.11.2025
|
| Schlagworte: | |
| ISSN: | 2093-4777, 2093-6931 |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | Purpose: This study aimed to evaluate the performance of an artificial intelligence (AI)-based analysis of uroflowmetry (UFM) curve images, enhanced with customized preprocessing techniques, to improve diagnostic accuracy for bladder outlet obstruction (BOO) and detrusor underactivity (DUA).
Methods: We retrospectively analyzed 2,579 UFM curve images from patients who underwent urodynamic study (UDS), including 725 normal and 1,854 abnormal cases (736 BOO and 1,387 DUA). A VGG16 convolutional neural network model was developed to perform 3 binary classification tasks: normal versus abnormal, BOO versus non-BOO, and DUA versus non DUA. To improve model performance, we implemented a preprocessing pipeline consisting of denoising, cropping, axis scaling, and color-coding of clinical parameters such as voided volume and postvoid residual volume (PVR). Model performance was evaluated using 5-fold stratified cross-validation and the area under the receiver operating characteristic curve (AUROC).
Results: Abnormal cases demonstrated a lower median maximum flow rate (8.9 mL/sec vs. 14.8 mL/sec), higher PVR (60.0 mL vs. 20.0 mL), and lower voiding efficiency (78.5% vs. 92.5%) than normal cases. Within the abnormal group, the BOO subgroup showed a higher PVR (80.0 mL) than the non-BOO subgroup (30.0 mL). After applying the preprocessing pipeline, model performance improved, with AUROC increasing from 0.807±0.024 to 0.827±0.016 for normal vs. abnormal classification, from 0.749±0.019 to 0.773±0.034 for BOO classification, and from 0.693±0.016 to 0.709±0.031 for DUA classification.
Conclusions: AI-based analysis of UFM curve images, enhanced through customized preprocessing, improved diagnostic accuracy in patients with lower urinary tract symptoms, effectively identifying BOO and DUA. This noninvasive method may serve as an adjunct or screening tool to reduce reliance on invasive UDS. KCI Citation Count: 0 |
|---|---|
| Bibliographie: | https://doi.org/10.5213/inj.2550266.133 |
| ISSN: | 2093-4777 2093-6931 |
| DOI: | 10.5213/inj.2550266.133 |