Bridging Fibrosis in liver
Embedded Files
Semantic Segmentation: Assigns each pixel in an image to a predefined class (e.g., epidermis, dermis, hair follicle, blood vessel). Here each color represents a different tissue type.
Semantic segmentation focuses on broader tissue types, while instance segmentation dives into individual objects.
Instance Segmentation: Identifies and outlines individual objects (e.g., nuclei, cells, glands) within each class. Each object is uniquely colored and outlined.
Instance segmentation provides more granular information, crucial for tasks like: Cell counting, Morphological analysis, Studying cellular interactions
Applications in Skin Histology:
Cancer diagnosis: Accurate segmentation of tumor cells and surrounding tissue aids in diagnosis and grading.
Disease analysis: Segmentation of specific structures (e.g., blood vessels, collagen fibers) helps assess disease severity and progression.
Drug development: Segmentation of target cells or structures enables evaluation of drug efficacy and potential side effects.
Image segmentation is the process of dividing an image into regions with similar properties. In histopathology, WSI image segmentation is used to identify and separate different tissue types, such as cancer cells, healthy cells, and stroma. This can be used for a variety of purposes, such as tumor classification, prognosis, and drug discovery.
Deep learning algorithms have been shown to be very effective at image segmentation, and they are increasingly being used in histopathology.
Here are some of the benefits of using deep learning for WSI image segmentation:
Accuracy: Deep learning algorithms have been shown to be more accurate than traditional methods for WSI image segmentation.
Speed: Deep learning algorithms can segment WSI images much faster than traditional methods.
Reproducibility: Deep learning algorithms are reproducible, meaning that they can be used to segment WSI images with consistent results.
Demo Test Run: Bridging Fibrosis segmentation
An automated machine learning lesion detection algorithm is tested as computational biomarker for NAFLD and NASH. Image shows bridging fibrosis ROI`s identified by the algorithm (marked bounding box with %age confidence) that the algorithm has never seen during training.
Hepatocellular Ballooning is an important pathology feature of NASH progression. An automated machine learning algorithm is able to detect the ballooning degeneration
The t-distributed stochastic neighbor embedding (t-SNE), a nonlinear visualization of the multidimensional histopathology imaging data into 3D space is tested in visualizing clinically relevant pathology features in NASH and other liver diseases. Most common pathology features of NASH such as steatosis, necrosis, fibrosis images patches are clustered in 3-D space by t-SNE. The algorithm learned to place fibrotic tiles with fat image tiles in the neighborhood of fibrosis and steatosis image tiles cluster. This technique could be helpful in quick review of NASH features on a whole slide histopathology liver images in estimating heterogeneity in inter-intra patient populations responding to drug.
Nuclei segmentation
Nuclei segmentation by multi-tissue segmentation Foundation model trained on CoNSeP dataset
The CoNSeP (Colorectal Nuclear Segmentation and Phenotypes) dataset is a valuable resource for histopathological image analysis. It consists of 41 H&E stained image tiles, each measuring 1,000 x 1,000 pixels at 40x magnification. These images were extracted from 16 colorectal adenocarcinoma (CRA) whole slide images (WSIs), each belonging to an individual patient. The dataset was scanned using an Omnyx VL120 scanner at the University Hospitals Coventry and Warwickshire, UK.
CoNSeP is widely used for tasks such as nuclear segmentation and classification, providing a benchmark for developing and evaluating algorithms in computational pathology. This dataset is crucial for advancing automated analysis techniques in colorectal cancer research, aiding in the development of more accurate diagnostic tools and treatment strategies.
Figures shown below are the examples of nuclei segmentation tested using foundation model
Page updated
Google Sites
Report abuse