Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster display session

388P - To identify circulating tumour cells by machine learning approach

Date

23 Nov 2019

Session

Poster display session

Topics

Targeted Therapy

Tumour Site

Presenters

Yuebin Liang

Citation

Annals of Oncology (2019) 30 (suppl_9): ix122-ix130. 10.1093/annonc/mdz431

Authors

Y. Liang, H. Lin, G. Tian, J. Yang

Author affiliations

  • Medical Department, Geneis Co. LTD, 100000 - Beijing/CN

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 388P

Background

More and more evidences suggest that circulating tumor cells (CTC) may be a good biomarker, not only as a prognostic indicator for tumors, but also for monitoring therapeutic effect and recurrence. However, the detection of CTC usually requires the final artificial judgment. This requires experienced pathologists and increases their workload. The application of machine learning in medical image recognition can effectively improve the level of automation and reduce the workload. So we hope to use machine learning to identify CTCs.

Methods

First, the python's openCV software package is used to segment the images of CTCs by image denoising, image filtering, edge detection, image expansion and contraction techniques. Secondly, the segmented cell images as a training set are trained using the CNN deep learning network. The CNN deep learning network includes input layer, intermediate hidden layer, and output layer. The middle hidden layer contains three layers, namely layer1, layer2 and lager3. Each intermediate hidden layer further includes convolution layer, excitation layer, and pooling layer. After the input layer, the cell images first enter the first intermediate hidden layer. The convolution layer of the first intermediate hidden layer is composed of 32 5x5 convolution kernels, which are then output to the pooling layer for dimension reduction through the ReLU excitation layer. After dimension reduction, the data is output from the first hidden layer to complete an entire feature extraction process. Then, through the second and third intermediate hidden layers in sequence, all feature extraction is completed. Finally, it enters the output layer and output the result, ie, CTCs or non-CTCs.

Results

We took 2920 cells from 732 patients for training and testing. Among them, 2000 cells were used as training set and 920 cells were used as testing set. The sensitivity and specificity of recognition reached 86.1% and 84.5%, respectively.

Conclusions

To identify CTC by machine learning can reach high sensitivity and specificity. We are further revising our methods of deep learning to achieve greater recognition effect.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.