AI-Powered Fetal Annotation by Logictive Solutions
Logictive Solutions collaborated with Deepecho, a leading medical technology company, on a critical annotation project focusing on identifying and labeling various fetal anatomical structures in ultrasound images. The annotation work involved using polygon, polyline, and ellipse tools to precisely outline structures such as the Myometrium, Ventricles, Choroid Plexus, Cerebellum, Cisterna Magna, Humerus, Heart, Lungs, AF Pocket, Placenta, and Spine. The project aims to create an AI model to track all the above aspects related to Fetal helping in radiologist / specific medical section to smoothen their day to day processes and analyst.
Services We Provided
Project Overview
Fetal Annotation based on Fetal Anatomy Ultrasound Annotation project was designed to support radiologists in identifying key fetal anatomical structures.
Target Scan Period
Ultrasound scan from 18 to 22 weeks of pregnancy.
Critical Role
Accurate identification of listed structures on fetal ultrasound plays a vital role in monitoring fetal development and detecting abnormalities early.
AI Training Requirement
Deepecho’s Fetal Ultrasound Image scanning AI systems required high-quality fetal annotated datasets to train their diagnostic AI models, making this project a crucial step in improving prenatal screening capabilities.
Key Project Objectives
The main objectives of the project were focused on precision, enablement, and standardization.
Produce Precise Annotations
To produce precise and consistent annotations of critical fetal anatomical structures in ultrasound images.
Enable AI Model Accuracy
To enable Deepecho’s AI models to learn and detect these structures accurately, thereby assisting radiologists in faster and more reliable diagnostics.
Standardize Annotation Protocol
To standardize the annotation process for fetal anatomy in the absence of pre-existing detailed guidelines.
Structured Project Approach
Guideline Development and Execution
Logictive Solutions adopted a structured, research-driven, and iterative process to ensure annotation accuracy and project success.
4.1 Guideline Development
Since no predefined annotation guidelines were provided, the team conducted extensive research to understand fetal anatomy visibility in ultrasound images. Preliminary guidelines were made based on research and medical expert guidelines and verification, where there were multiple communication session with the radiologists(client).
4.2 Annotation Execution
Using specialized annotation tools, our team annotated each ultrasound image, focusing on clarity, boundary accuracy, and structure differentiation.
Structured Project Approach
Verification and Data Structuring
4.3 Expert Verification
The guidelines and annotated images were then reviewed and validated through consultations with medical experts/radiologists(client) to ensure anatomical accuracy. Feedback loops were established to refine the guidelines and standardize labeling practices across the dataset.
4.4 Data Structuring
Finally, the annotated outputs were converted from unstructured image data into structured formats compatible with Deepecho’s AI systems. This included proper labeling conventions, metadata alignment, and quality assurance checks.
Annotation Tool Application
Different CVAT annotation tools were applied depending on the anatomical structure’s shape to ensure maximum precision during the Fetal Annotation Execution phase.
Polygon Tool
Used for irregular structures like the placenta and lungs.
Polyline Tool
Used for linear structures such as the spine (Initial Stage | Currently Discontinued).
Ellipse Tool
Used for rounded organs like ventricles and cisterna magna.
Project Challenges
To address these challenges, Logictive Solutions implemented several effective strategies:
Lack of Pre-Established Guidelines
There were no standard annotation protocols for fetal ultrasound imagery provided at the start.
Absence of Immediate Medical Expertises
The annotation team initially lacked direct access to radiologists(client), making it difficult to identify less prominent anatomical structures.
Complexity of Ultrasound Images
Ultrasound images often have low contrast and high variability, requiring careful interpretation to avoid mislabeling.
Challenge Mitigation Strategies
To address these challenges, Logictive Solutions implemented several effective strategies:
Guideline Creation through Research
The team developed initial annotation guidelines by studying medical imaging literature and referencing fetal anatomy charts.
Medical Expert Consultations
Initially, regular consultations with radiologists(client) were arranged to review ambiguous cases and validate the evolving guidelines.
Iterative Training and Feedback
Fetal Annotation team members underwent targeted training sessions based on expert feedback, which significantly improved accuracy and consistency over time.
Outcomes and Results
The project was successfully completed with high-quality annotations delivered on time. Key outcomes included:
- Accurate and consistent Fetal annotation
of thousands of ultrasound images.
- Improved dataset quality
that enabled Deepecho to enhance the performance of their Fetal AI diagnostic models.
- Standardized fetal anatomy annotation protocol
that can be reused for future medical imaging projects.
- Positive client feedback
highlighting the team’s ability to work without initial guidelines and still deliver medically validated results.
Conclusion and Future Improvements
The Fetal Anatomy Ultrasound Annotation project demonstrated Logictive Solutions’ ability to handle complex medical imaging tasks with precision and adaptability. By combining research, structured workflows, expert consultations, and technical rigor, the team delivered a dataset that contributed to advancing AI-driven prenatal care — ultimately supporting radiologists in ensuring better maternal and fetal health outcomes.
Lessons Learned / Future Improvements
- Importance of Clear Guidelines:
Developing structured annotation guidelines early in the project can significantly improve team efficiency and reduce revisions.
Scalability of Structured Processes: The framework established in this project can be replicated and scaled for other medical imaging tasks, such as obstetric, cardiac, or organ-specific datasets.