Quantitative assessment of new MRI-based measurements to differentiate low and high stages of pelvic organ prolapse using support vector machines

Multi-label classification of pelvic organ prolapse using stress magnetic resonance imaging with deep learning

INTRODUCTION AND HYPOTHESIS

We aimed to develop a deep learning-based multi-label classification model to simultaneously diagnose three types of pelvic organ prolapse using stress magnetic resonance imaging (MRI).

METHODS

Our dataset consisted of 213 midsagittal labeled MR images at maximum Valsalva. For each MR image, the two endpoints of the sacrococcygeal inferior-pubic point line were auto-localized. Based on this line, a region of interest was automatically selected as input to a modified deep learning model, ResNet-50, for diagnosis. An unlabeled MRI dataset, a public dataset, and a synthetic dataset were used along with the labeled image dataset to train the model through a novel training strategy. We conducted a fivefold cross-validation and evaluated the classification results using precision, recall, F1 score, and area under the curve (AUC).

RESULTS

The average precision, recall, F1 score, and AUC of our proposed multi-label classification model for the three types of prolapse were 0.84, 0.72, 0.77, and 0.91 respectively, which were improved from 0.64, 0.53, 0.57, and 0.83 from the original ResNet-50. Classification took 0.18 s to diagnose one patient.

CONCLUSIONS

The proposed deep learning-based model were demonstrated feasible and fast in simultaneously diagnosing three types of prolapse based on pelvic floor stress MRI, which could facilitate computer-aided prolapse diagnosis and treatment planning.

Feasibility of a deep learning-based method for automated localization of pelvic floor landmarks using stress MR images

INTRODUCTION AND HYPOTHESIS

Magnetic resonance imaging (MRI) plays an important role in assessing pelvic organ prolapse (POP), and automated pelvic floor landmark localization potentially accelerates MRI-based measurements of POP. Herein, we aimed to develop and evaluate a deep learning-based technique for automated localization of POP-related landmarks.

METHODS

Ninety-six mid-sagittal stress MR images (at rest and at maximal Valsalva) were used for deep-learning model training and generalization testing. We randomly split our dataset into a training set of 73 images and a testing set of 23 images. One soft-tissue landmark (the cervical os [P1]) and three bony landmarks (the mid-pubic line [MPL] endpoints [P2&P3] and the sacrococcygeal inferior-pubic point [SCIPP] line endpoints [P3&P4]) were annotated by experts. We used an encoder-decoder structure to develop the deep learning model for automated localization of the four landmarks. Localization performance was assessed using the root square error (RSE), whereas the reference lines were assessed based on the length and orientation differences.

RESULTS

We localized landmarks (P1 to P4) with mean RSEs of 1.9 mm, 1.3 mm, 0.9 mm, and 3.6 mm. The mean length errors of the MPL and SCIPP line were 0.1 and -2.1 mm, and the mean orientation errors of the MPL and SCIPP line were -0.7° and -0.3°. Our method predicted each image in 0.015 s.

CONCLUSIONS

We demonstrated the feasibility of a deep learning-based approach for accurate and fast fully automated localization of bony and soft-tissue landmarks. This sped up the MR interpretation process for fast POP screening and treatment planning.