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 viewing and lunch

179P - Developing the AI program for TNBC subtyping

Date

12 May 2023

Session

Poster viewing and lunch

Presenters

EBRUCAN BULUT

Citation

Annals of Oncology (2023) 8 (1suppl_4): 101222-101222. 10.1016/esmoop/esmoop101222

Authors

E. BULUT1, R.F. Balaban1, N. Huriyet1, M.M. Önal2, U. Unal3, H. Tezcan1, G. Gokalp1, U. Egeli1, V. Polatkan1, M.S. Gokgoz1, G. Çecener1

Author affiliations

  • 1 Uludag University - Faculty of Medicine, Bursa/TR
  • 2 Yildiz Technical University, Bursa/TR
  • 3 Uludag University - Faculty of Science, Bursa/TR

Resources

Login to get immediate access to this content.

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

Abstract 179P

Background

Triple-negative breast cancer (TNBC) is a heterogeneous disease with poor prognosis and can be classified into different molecular subtypes. There is a need to develop a computer-aided diagnosis system with artificial intelligence (AI) techniques to increase the accuracy of evaluating suspicious breast lesions and identifying subtypes.

Methods

A multi-gene analysis panel consisting of 14 genes and 2 reference genes was designed from the TCGA, and cBioportal datasets, which were determined in accordance with in silico analyses in 488 TNBC patients (TCGA:123, cBioportal:365). In the AI part of the study, a convolutional neural network model with 1,838,915 parameters was trained to classify Ultrasound (US) images as “normal, benign, malignant.” In the model, the images of the patients were trained with 1000 epochs.

Results

Expression levels of related genes in paraffin-embedded tumors and normal tissues of 38 patients with TNBC were investigated by the RT-PCR method. When the gene expression differences of the tumor and normal tissues of the patients were compared, AR (p=0.013), ER (p=0.025), PGR (p=0.007), FOXA1 (p=0.0154), CXCL11 (p=0.0037), IDO1 (p=0.0005), ADH1B (p=0.018), ADIPOQ (p<0.0001) and CHAD (p=0.0017) genes were found to have a statistically significant difference in expression compared to normal tissue. While no significant expression difference was detected in FOXC1 (p=0.3914), CXCL13 (p=0.2138), LAMP3 (p=0.2576), FGFR2 (p=0.077) and DHRS2 (p=0.715) genes. As a result, it was found that 68.5% of the patients showed a single subtype-specific expression profile (10 BLIA, 3 BLIS, 1 MES, and 12 LAR subtypes), and 31.5% of the patients showed an expression profile of more than one subtype. A random forest model was chosen among the trained models, and in the first stage, 81% accuracy was obtained in the subtyping of TNBC and 83% in the classification of US images.

Conclusions

It sheds light on the heterogeneous structure and the role of molecular subtyping in the tumorigenesis of TNBC patients and may contribute to routine clinical practice and the regulation of targeted therapy protocols.

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.