BiliQML
AI consultant and developer of comprehensive computational workflow
BiliQML is a supervised machine learning platform designed to quantify multiple biliary forms (such as immature bile ducts, ductal plates, and mature bile ducts) from large, whole-slide histopathological images of mouse liver tissues. This tool addresses key challenges in existing liver research, where traditional histological methods are limited by low reproducibility and time-intensive manual quantification. BiliQML combines image analysis and machine learning techniques to automatically identify and classify cholangiocyte groupings across a variety of disease models, offering a scalable, robust, and unbiased method for biliary form quantification.
Some of the key challenges I addressed as a part of this project include:
1. Training Dataset Preparation: The training dataset consisted of large amounts of manually labeled tiles of stained liver tissue, each categorized into one of three biliary forms or a non-classification label. It was a challenge to generate an adequately diverse and representative dataset of both healthy and diseased liver samples to train the model.
2. Image Preprocessing and Feature Extraction: Whole-slide images were annotated and processed using advanced computer vision methods. Novel features such as area, eccentricity, solidity, and Euler number were computed for each biliary form. Feature selection was optimized to prevent overfitting and reduce model bias.
3. Machine Learning Algorithm Selection: Multiple classification algorithms were tested to find the optimal model for biliary form classification.
4. Prediction Accuracy and Model Testing: The performance of BiliQML was validated across seven cholangiopathy mouse models, including genetic, toxicological, surgical, and therapeutic interventions to ensure that the model was sufficiently broad.
5. Visualization and Spatial Analysis: In addition to classification, BiliQML allows researchers to visualize the spatial distribution of biliary forms within the original whole-slide images.
Skills developed include: machine learning, AI, histology, in vivo mouse models, hepatology, image segmentation, feature engineering, spatial visualization, model validation