Work

Development and external validation of deep learning models for spontaneous preterm birth prediction from mid-trimester cervical ultrasound

Year: 2026
Type: article
Study design: Observational Study
Abstract: Preterm birth is the leading cause of neonatal death. Despite sustained efforts to identify high-risk women in the mid-trimester, accurate prediction remains difficult. Quantitative cervical ultrasound texture has been proposed as a predictor of spontaneous preterm birth. However, earlier models were developed in small single-centre samples and were not externally validated. We developed image-texture (Local Binary Patterns with a Random Forest), deep-learning (Vision Transformer), clinical-variable, and multimodal models to predict spontaneous preterm birth on the prospective GARBH-Ini cohort. We then externally validated our best models on an independent cohort scanned on a different ultrasound machine. Our best overall model reached an internal-test area under the receiver-operating-characteristic curve of 0.71 (95% CI 0.59, 0.83), but performed modestly at 0.52 (95% CI 0.38, 0.67) externally. The deep-learning and multimodal models did not perform better. Discrimination appeared higher in a clinically high-risk subgroup at the 34-week threshold. These estimates were imprecise because of few cases and are hypothesis generating which need to be confirmed in future studies. Among the several likely reasons for the modest external performance is the heterogeneity of preterm birth. Predicting distinct preterm-birth subtypes separately, and integrating additional biomarkers and data domains, might improve model performance.
Authors: Radhika Chanian, Divyanshu Mishra, Rahul Jain, Nikhil Sharma, Ashok Khurana, Reva Tripathi, Abhinav Jain, GARBH-Ini study group, Shinjini Bhatnagar, Vineeta Bal, Bhabatosh Das, Mahadev Dash, Bapu Koundinya Desiraju, Pallavi Kshetrapal, Sumit Misra, Uma Chandra Mouli Natchu, Satyajit Rath, Kanika Sachdeva, Dharmendra Sharma, Amanpreet Singh, Shailaja Sopory, Ramachandran Thiruvengadam, Nitya Wadhwa, Arindam Maitra, Partha P. Majumder, Souvik Mukherjee, Tushar K. Maiti, Monika Bahl, Shubra Bansal, Umesh Mehta, Sunita Sharma, Brahmdeep Sindhu, Sugandha Arya, Rekha Bharti, Harish Chellani, Pratima Mittal, Anju Garg, Siddharth Ramji, Reva Tripathi, Alpesh Goyal, Yashdeep Gupta, Smriti Hari, Nikhil Tandon, Rakesh Gupta, Dinakar M. Salunke, G.Balakrish Nair, Gagandeep Kang, Nitya Wadhwa, J. Alison Noble, Ramachandran Thiruvengadam, Bapu Koundinya Desiraju, Shinjini Bhatnagar
Language: English
Field: Medicine
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7169712518
FWCI:
Cites: 23
Cited by:
Related to: 10