A study of multi-task learning using a VoVNet-OSA block enhanced U-Net on the Med++ MNIST dataset

SYUAN-HAO LI

Abstract:

In this study, we propose a multi-task learning framework using a modified VoVNet-OSA Block Enhanced UNet, named
VovUnet_Var, for image segmentation and classification on the Med++ MNIST dataset. VovUnet_Var features downsampling
(DownOsa) and upsampling (UpOsa) blocks, with a classification head. The architecture sequentially downscales the input
image to capture hierarchical features and uses adaptive average pooling and a fully connected layer for classification. For
segmentation, upsampling layers restore spatial dimensions, producing segmentation masks. This model effectively handles
complex medical imaging tasks, providing a robust solution for simultaneous image segmentation and classification.