Job Detail

Master’s student position

Positive-unlabeled learning for tumor region localization

Ref. 2019-29

Job Description

In the IBM Research Computational Pathology team, one of our main points of focus is analyzing digitized histopathology images.

Tissue biopsies are stained and digitized to produce whole-slide images (WSIs). A WSI is a gigapixel-sized image that is captured at multiple magnifications. Pathologists analyze and estimate the stage and grade of cancer in WSIs for diagnosis and prognosis purposes. Tumor region detection is an important measure for this assessment process. Detection of tumor regions confines the analysis region and is used for other quantification metrics such as tumor size, distance to resection margin etc. However, delineation of tumor regions in a clinical setting is usually a tedious manual and time-consuming task that is prone to observer variability.

Deep-learning techniques can serve to automate tumor detection. Several supervised deep-learning solutions have been proposed to address the task, but they require large accurate tumor region annotations, which is expensive and cumbersome to acquire. However, highly confident high-power fields (HPFs), located on the periphery of invasive tumor regions, can be obtained from pathologists, which can be used for identifying invasive tumor regions. HPFs can only provide information about the appearance of s tumor, thus making the task a positive-unlabeled (PU) learning problem, whereas tumors within an HPF represent the positive class and the rest is unlabeled.

P. Pati et al.,
“Deep positive-unlabeled learning for region of interest localization in breast tissue images,”
SPIE-Medical Imaging, 2018.

J. Bekker et al.,
“Learning from Positive and Unlabeled Data: A Survey,”
arXiv:1811.04820, 2018.

K. Sirinukunwattana et al.,
“Improving Whole Slide Segmentation Through Visual Context-A Systematic Study,”
MICCAI, 2018.

Goals of the Master’s thesis:

• Literature review of tumor region detection approaches.


• Literature review of PU learning and Deep clustering algorithms.


• Develop a context-aware deep clustering-based framework for a PU-learning tumor region detection task. Potentially extend the framework to understand tumor heterogeneity.


• Perform ablation study and evaluation on large WSI dataset.

Requirements

• Working knowledge of Python.


• Strong theoretical understanding of classical computer vision techniques, machine-learning and deep-learning architectures.


• Practical experience with Keras, Tensorflow or PyTorch deep-learning frameworks.

Diversity

IBM is committed to diversity at the workplace. With us you will find an open, multicultural environment. Excellent, flexible working arrangements enable both women and men to strike the desired balance between their professional development and their personal lives.

How to apply

If you are interested, please send your application to Antonio Foncubierta (fra@zurich.ibm.com) including your CV and a motivation letter.

Preferable starting date: August – October 2019.

Note: The position is only for Master’ thesis students. Internships are not available at the moment.

Connect with us

Downloads

• Overview presentation of IBM Research – Zurich Lab (7 MB)


• IBM Research – Zurich Lab Fact Sheet (1.5 MB)


• IBM’s Diversity brochure (168 KB)