XClose

UCL Centre for Medical Image Computing

Home
Menu

Morgan Roberts & Cameron Shand – CMIC/WEISS joint seminar series

24 March 2021, 1:00 pm–2:00 pm

Morgan Roberts & Cameron Shand a talk as part of the CMIC/WEISS joint seminar series

Event Information

Open to

All

Organiser

cmic-seminars-request@cs.ucl.ac.uk

Speaker: Morgan Roberts

Title: Ultrasound Tomography for Breast Cancer Screening – Lowering the Barrier to Entry.

Abstract: Ultrasound Tomography (UST) is an emerging imaging modality for breast cancer screening that generates quantitative sound speed, density and absorption maps in 3D. It is non-ionising and requires no breast compression, making it a promising alternative to the mammogram. However, data acquisition and reconstruction times are currently too long to be clinically useful. To address this, work on novel reconstruction methods often requires data to be collected with new hardware configurations. However, researchers in the UST community face a high barrier to entry, since UST hardware is not available off the shelf, and custom systems have a high cost and long lead time.

In this talk, I will present the relationship between image reconstruction and data acquisition in UST, and the work we are doing to develop an open-source modular UST system. This system allows end-users to fabricate and modify UST sensor arrays to be compatible with different reconstruction techniques, and allows novel array configurations to be tested.

 

Speaker: Cameron Shand

Title: Evolving Challenging Synthetic Clusters (or Why evolutionary algorithms are still cool)

Abstract: Due to the inherently limited perspective clustering algorithms take on cluster structure, and the difficulty in evaluating their performance, synthetic data plays a vital role in the validation and development of these algorithms. Synthetic clusters are often, however, too contrived or lack sufficiently defined properties to provide algorithmic insights. To this end, we developed HAWKS, an evolutionary framework for evolving datasets. This framework currently has two separate approaches: evolving datasets to a specified level of difficulty (according to a cluster index); or, evolving to maximize the performance difference between clustering algorithms directly. Characterisation of datasets (from HAWKS and other popular generators/repositories) is enabled through the construction of an instance space using properties specific to clustering. In this talk, I'll show how HAWKS is useful to better understand clustering algorithms (for research or even teaching), and why evolutionary algorithms are still cool (if you had any doubt!).

 

Chair: Neil Oxtoby