PROJECT SUMMARY/ABSTRACT Interpretation of fetal brain MRI continues to be a challenge due to fetal motion even though single-shot techniques such as T2 weighted Half Fourier Single-shot Turbo spin-Echo (HASTE) acquisitions are used. Unpredictable fetal movement causes artifacts and images acquired in oblique orientations. As a result, the radiologist must interpret multiple stacks of imperfect HASTE images in varying oblique orientations, visually interrogating up to 1,000 independent 2D images. Additionally, multiple linear measurements are used to assess brain development. This approach is time-consuming and mentally taxing. Ironically, methods to create coherent fetal brain volumes have existed for over a decade in the research world but none have transitioned into clinical workflows due to long reconstruction times, the need for specialized hardware, and no infrastructure to deposit the reconstructed volumes/biometrics into the picture archiving and communication system (PACS) for a radiologist’s review. We will leverage the latest deep-learning strategies to perform rapid, robust, and accurate fetal brain reconstructions and automatically generate fetal brain biometrics. We will build on the unique open- source Children’s Hospital Research Integration System (ChRIS) to integrate results into the clinical workflow. In Aim 1 we will develop fast methods for reconstructing coherent 3D fetal brain volumes from a collection of clinically acquired 2D HASTE slice stacks, providing accurate automated metrics to support clinical interpretation. Our deliverable is an algorithm that i) successfully performs fetal brain reconstructions in >95% of cases in under 2 min and ii) provides important brain biometrics within 1 standard deviation of expert human measures. In Aim 2 we will integrate fetal brain reconstructions and derived metrics into the clinical workflow using ChRIS, within 5 min of study completion. In Aim 3 we will assess the impact of coherent fetal brain volumes and automated metrics on radiological interpretation. We hypothesize that expert fetal brain interpretations will be faster, more sensitive, specific, and concordant in the detection of abnormalities when viewing reconstructed volumes and using automated metrics compared to viewing 2D HASTE stacks as acquired. Ground truth will be provided by neonatal brain MRIs and the cohort will be enriched with cases where neonatal MRIs discovered findings missed on fetal MRIs. The deliverable is a rigorous quantitative evaluation of the impact of coherent volume reconstruction and automatic brain biometrics on the efficacy of radiological interpretation of fetal MRI. If successful, this project will enable radiologists to interpret fetal brain MRIs as one volume instead of numerous stacks of images in varying obliquities, saving time, and increasing accuracy. The automated metrics will further enrich the information available to radiologists without requiring additional physician time. The ChRIS infrastructure is freely available on GitHub and will support the next phase of dissemination of this and other innovations to interested institutions for integration into their clinical workflows.

PROJECT NARRATIVE Interpretation of fetal brain MRIs remains challenging due to unpredictable fetal motion. We will address this challenge by developing and integrating a rapid and robust super-resolution slice-to-volume reconstruction (SVR) into the clinical workflow. If successful, this project will enable radiologists to interpret fetal brain MRIs as one volume instead of numerous stacks of images in varying obliquities, saving time and increasing accuracy.