Hybrid machine learning method for support brain tumor diagnosis

Congress:

ECR 2020

Poster Number:

C-07053

Type:

Scientific Exhibit

Keywords:

Multicentre study, Experimental, Retrospective, Cancer, Segmentation, Computer Applications-Detection, diagnosis, Neural networks, MR-Diffusion/Perfusion, MR, Neuroradiology brain, MR physics, Artificial Intelligence, Imaging Informatics

Authors:

I. Stathopoulos¹, L. Serio¹, U. Gentile¹, E. Karavasilis², V. G. Syrgiamiotis³, G. Velonakis², E. Efstathopoulos⁴; ¹Geneva/CH, ²Athens/GR, ³GOUDI/GR, ⁴Alimos/GR

DOI:

10.26044/ecr2020/C-07053

DOI-Link:

https://dx.doi.org/10.26044/ecr2020/C-07053

Table 1: Images per Modality

Table 2: Images per Class

Methods and materials

The proposed tool is based on a customized Faster R-CNN[9] deep learning model based on the VGG-16 architecture[10] for tumour detection and classification and the Fuzzy C-means (C=3) clustering algorithm for segmentation of the detected tumor part.

A set of 234 multimodal axial - T1-weighted (T1), T2-weighted (T2), Fluid-Attenuated Inversion Recovery (FLAIR), diffusion weighted images (DWI), contrast-enhanced T1-weighted (T1ce)- MRIs, labelled by two radiologists, was used for the training phase. Table 1

All images were abnormal belonging to two classes: “tumor” and “other pathologies” (“no-tumor”). Table 2

The purposed tool detects the abnormality in the image, crops it, classifies it and segments it to three different regions (C=3).