The UCSB Bio-Segmentation Benchmark dataset consists of 2D/3D images (Section 1) and time-lapse sequences that can be used for evaluating the performance of novel state of the art computer vision algorithms. Tasks include segmentation, classification, and tracking.

For each class of problem, at least one ground truth dataset is available. We also provide performance metrics for comparing the results of the algorithms with the ground truth. Additional ground truth data will be posted as they become available. Matlab code for performance evaluation is also available. If you use this dataset in your research please cite our ICIP'08 paper (see the citation below) in your publications.

Current results using methods developed at UCSB are provided in the tables below. If you would like us to post your results, please upload them at http://ganga.ece.ucsb.edu:8080 (Section 2). In addition, we would be happy to post new datasets if made available.

The data is organized in 2 different ways,one based on image content type (subcellular, cellular and tissue level data) and the other one is based on the image processing methodology (segmentation or classification or tracking).

Summary Table

Datasets

Algorithms and Performance Evaluation

Section 1

By image content type:

Subcellular level

Microtubule dataset

Cellular level

Photoreceptors in 2D Retinal Images

For example, in retinal images, the number of photoreceptor nuclei in the outer nuclear layer (ONL), depicted in this image, is one of the important
measurements of the retina degeneration. UCSB retinal dataset consists of 40 laser scanning confocal images of normal and 3-day detached feline retinas (20 normal
and 20 3-day detached). The detached retinal samples are obtained by surgically detaching a retina and leaving the animal in the detached retinal state for a
certain period of time before imaging the tissue samples. Images were collected using a laser scanning confocal microscope from tissue sections. For each image,
the ground truth, consists of an ONL binary mask and the corresponding manual cell count in the ONL layer by three different experts.

Breast Cancer Cell

There are about 50 H&E stained histopathology images used in breast cancer cell detection with associated ground truth data available. Routine histology uses the stain combination of hematoxylin and eosin, commonly referred to as H&E. These images are stained since most cells are essentially transparent, with little or no intrinsic pigment. Certain special stains, which bind selectively to particular components, are be used to identify biological structures such as cells. In those images, the challenging problem is cell segmentation for subsequent classification in benign and malignant cells. The ground truth have been obtained for one image containing benign cells.

COS1 cells

COS1 cells (immortalized African monkey kidney cells) are collected through confocal microscopy imaging. The images are of both wild-type COS1 cells (non-transfected) and tau transfected COS1
cells and these cells were imaged at 7 different timepoints after treatment (2hrs, 6 hrs, 12 hrs, 48 hrs, 72, hrs, and 120 hrs). The challenge in these images is to identify dead (red), alive (blue)
and total number of cells (blue in green background). Ground truth has also been collected for this dataset and is represented by binary masks.

Photoreceptor in 3D Retinal Images

Tissue level

Retina Layers (regions)

Retinal Boundaries

Confocal microscope images of retinas taken during detachment experiments are critical components for understanding the structural and cellular changes of a retina in response to disease and injury. As the first step of any other analysis (e.g. before cell counting), it is crucial to have a reliable map of the retinal layers. Hundreds of retinal images and layer ground truth are part of the benchmark. Four major layers of the retina are segmented manually: the ganglion cell layer (GCL), the inner nuclear layer (INL), the outer nuclear layer (ONL), and the outer segments (OS).

Section 2

How to evaluate the segmentation results

The user has different ways to evaluate the performance of their algorithms. The choices are:

• download the Matlab code to self-evaluate the algorithm performance,
• use web-based evaluator: http://ganga.ece.ucsb.edu:8080 to upload the analysis results in the correct format and run the web-based evaluator, the evaluation results will be automatically displayed on the web site.
  

Work based on the dataset should cite our ICIP '08 paper:

@inproceedings{Drelie08-298,

author = {Elisa Drelie Gelasca and Jiyun Byun and Boguslaw Obara and B.S. Manjunath},

title = {Evaluation and Benchmark for Biological Image Segmentation},

booktitle = {IEEE International Conference on Image Processing},

location = {San Diego, CA},

month = {Oct},

year = {2008},

url = {http://vision.ece.ucsb.edu/publications/elisa_ICIP08.pdf}}

Please Note: To ensure the integrity of results on the test data set, you may use the images and human segmentations in the training set for tuning your algorithms but your algorithms should not have access to any of the data (images or segmentations) in the test set until your are finished designing and tuning your algorithm.

Contact

Kristian Kvilekval ( This e-mail address is being protected from spambots. You need JavaScript enabled to view it ), B.S. Manjunath ^1,2

¹Department of Electrical and Computer Engineering, ²Center for Bio-image Informatics, University of California, Santa Barbara, CA.