UCL Centre for Medical Image Computing


Daniele Ravi - Kevin Bronik - CMIC seminar series

30 January 2019, 1:00 pm–1:45 pm

Talks by Daniele Ravi and Kevin Bronik, Centre for Medical Image Computing

Event Information

Open to





Roberts LT 106
Roberts Building
Malet Place

Daniele Ravi - Adversarial training with cycle consistency for  unsupervised super-resolution in endomicroscopy


In recent years, endomicroscopy has become increasingly used for diagnostic purposes and interventional guidance. It can provide intraoperative aids for real-time tissue characterization and can help to perform visual investigations aimed for example to discover epithelial cancers.
Due to physical constraints on the acquisition process, endomicroscopy images, still today have a low number of informative pixels which hampers their quality. Post-processing techniques, such as Super-Resolution (SR), are a potential solution to increase the quality of these images. SR techniques are often supervised, requiring aligned pairs of low-resolution (LR) and high-resolution (HR) images patches to train a model. However, in our domain, the lack of HR images hinders the collection of such pairs and makes supervised training unsuitable. For this reason, we propose an unsupervised SR framework based on an adversarial deep neural network with a physically-inspired cycle consistency, designed to impose some acquisition properties on the super-resolved images. Our framework can exploit HR images, regardless of the domain where they are coming from, to transfer the quality of the HR images to the initial LR images. This property can be particularly useful in all situations where pairs of LR/HR are not available during the training. Our quantitative analysis, validated using a database of 238 endomicroscopy video sequences from 143 patients, shows the ability of the pipeline to produce convincing super-resolved images. A Mean Opinion Score (MOS) study also confirms this quantitative image quality assessment.


Kevin Bronik - title:  Analyzation and optimization of the existing open source codes for multiple sclerosis white matter lesion segmentation


Multiple sclerosis (MS) is a lifelong condition, a result in brain and/or spinal cord disfunction that in most cases causes serious disability, a condition that is difficult to get a correct diagnosis on and to say whether the symptoms are caused by multiple sclerosis (MS) at first or something else, because in some cases it could be similar to other conditions as well. There is no quick/accurate medical diagnosis for MS which could be done through a single test, therefore, prior to any accurate diagnostics that confirms MS, several tests such as Neurological examination, MRI scan, Lumbar puncture and Blood tests need to be done to determine positively symptoms related to MS such as problems with vision, arm or leg movement etc. There are several existing supervised and unsupervised learning algorithm designed for lesion detection using MRI scans. The current study has been investigating the performance of three open source codes namely: BIANCA, LST and NICMSLESION. Comparison the results of commonly accepted performance analysis reveals the algorithm used in NICMSLESION for multiple sclerosis white matter lesion segmentation, based on convolutional deep neural networks, is the most reliable lesion segmentation algorithm of all. The current challenges of the study are the progress of multi-criterion optimizations such as getting better agreement between manual and predicted lesion masks and reducing computational complexity using NICMSLESION, where introducing new framework such as Horovod into NICMSLESION will help to reduce the complexity and make distributed Deep Learning faster.


Other events in this series