Discover the advanced hybrid decision model, medical color science, and safety frameworks powering LiverPoop's screening engine.
The LiverPoop platform was designed to support early screening of potentially abnormal stool pigmentation patterns associated with pediatric liver conditions such as neonatal cholestasis.
Instead of relying on a single black-box AI model, the system uses a hybrid explainable architecture that combines multi-layer analytical models, color science analysis, image processing workflows, and medical safety logic.
The objective of the platform is to provide explainable, robust, and clinically safer screening-support outputs under real-world mobile imaging conditions.
The core screening analysis is driven by our hybrid decision model, which processes multiple analytical layers operating together inside a hybrid medical decision engine.
Verifies stool color against reference datasets of medically reviewed color ranges.
Measures perceptual distance between uploaded samples and clinically validated references.
Computes probability for abnormal (pale/acholic) or normal (pigmented) categories.
Identifies texture variations, spatial visual patterns, and mixed stool coloration.
The outputs from these layers are processed using a hybrid medical safety engine before generating the final screening result. This multi-layer architecture improves screening safety, explainability, and real-world reliability.
Before any analysis is performed, uploaded images undergo multiple preprocessing stages to ensure consistency:
Region of Interest (ROI) selection to focus on the stool sample
Resolution resizing and optimization for consistent analysis
Lighting normalization and brightness stabilization to reduce variability from flash, shadows, or camera sensors
To prevent inaccurate outputs, the platform includes a dedicated image quality assessment layer. The system evaluates factors such as excessive blur, low brightness, glare, or poor image clarity.
If image quality is insufficient to guarantee a reliable screening, the system outputs a "Retake Recommended" message rather than forcing an uncertain classification — ensuring patient safety above all else.
One of the primary goals of the platform is maintaining transparency. Unlike fully opaque, black-box AI systems, the LiverPoop architecture allows validation of the color similarity behavior, decision rule interactions, image quality limitations, and reference matching logic. This improves clinical interpretability and supports validation workflows.
The platform uses secure, scalable cloud-based infrastructure and encrypted database services. This backend infrastructure supports real-time analysis, secure processing, and data integrity while keeping sensitive user records protected and confidential.
Developed using secure, cross-platform mobile frameworks for smooth UI performance, ROI selection, and multi-language support.
Developed using robust processing languages and specialized image processing libraries for secure validation, feature extraction, and model inference.
The platform was developed and evaluated using a comprehensive training dataset of stool image samples. This dataset includes real-world mobile images capturing lighting variability, different cameras, and multiple pigmentation patterns to ensure generalizability and robustness under practical screening conditions.
For technology, research, or general inquiries, please reach out to: support@liverpoop.com