Problem
Research questions and friction points this paper is trying to address.
Uterine contractions
Fetal MRI
Uterine Contractile Activity Detection
Weakly-supervised learning
STORK: Spatio-Temporal Observation of uterine contRactions via neural networKs
🤖 AI Summary
This study addresses the challenge that identifying uterine contractions in fetal MRI relies on manual assessment and lacks fine-grained annotations by proposing a weakly supervised multiple instance learning framework. The method pioneers the decomposition of 3D spatiotemporal convolutions into parallel temporal hyperplane branches, integrated with Demons displacement field fusion and linear mean pooling. This design captures tissue motion consistency while circumventing the computational overhead of full 4D processing, enabling the recovery of frame-level contraction scores from sequence-level labels alone. Evaluated on approximately 700 dynamic fetal MRI cases, the model achieves an AUROC of 95.0%, significantly outperforming existing baselines. Ultimately, this work facilitates automated, fine-grained phenotypic analysis of uterine behavior.
📝 Abstract
Uterine contractions in fetal MRI are typically identified manually and discarded, limiting insights into contraction dynamics. We formalize Uterine Contractile Activity Detection (UCAD) as a weakly-supervised learning problem and introduce STORK, a multi-instance learning model trained on dynamic MRI series using only coarse, series-level labels. STORK factorizes 3D spatio-temporal convolutions into parallel branches across temporal hyperplanes to capture coherent tissue motion without the cost of full 4D convolutions. Per-frame embeddings, combining intensity and Demons-estimated displacement fields, are aggregated by a linear mean-pooling head. This ensures that frame-level contraction scores can be recovered post-hoc without frame-level training supervision. Evaluated on around 700 multi-vendor dynamic fetal MRI series, STORK achieves a series-level AUROC of 95.0% and AUPRC of 94.6%, substantially outperforming 3D ResNet and ConvNeXt baselines. Grad-CAM analysis suggests that the model draws on predictive features extending beyond the placenta into the uterine tissue, offering an automated tool for richer phenotyping of uterine behavior.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Weakly-supervised learning
Multiple instance learning
Spatio-temporal convolution factorization
Displacement field fusion
Uterine contractile activity detection
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
No related jobs found.
