Project: Multi-task learning for image classification of renal tumors

Status

Completed (2020-2023)

Type

Master’s

Team

• Alexandre Ayotte¹ (2020-today)
• Martin Vallières¹ (2020-today)

¹ Computer science department, Université de Sherbrooke, Sherbrooke (QC), Canada

Data

For this project, data from 1,082 patients from 5 institutions with clinical data such as age, sex and tumor size, as well as 3D MRI images (T1, T2) with the region of interest of the tumors are used.

Classification

From the previously stated data, the goal is to develop a multi-task model to perform three binary classification tasks, which are:

• Malignancy classification (399 benign, 683 malignant)
• Subtype classification (only for malignant tumors) (158 papillary, 441 clear cell)
• Grade classification (only for malignant tumors) (391 low grade, 202 high grade)

Objectives

• Develop an integrated decision support system for the three classification tasks
• Evaluate the effectiveness of multi-task learning in the context of medical imaging and tumor classification
• Establish the benefit of radiomics as an auxiliary task in the context of multi-task learning