How Much Data are Enough? Investigating Dataset Requirements for Patch-Based Brain MRI Segmentation Tasks

• DOI: 10.48550/arxiv.2404.03451
• Publication Date: 2024-04-04
• Authors: Dongang Wang; Peilin Liu; Hengrui Wang; Heidi Beadnall; Kain Kyle; Linda Ly; Mariano Cabezas; Geng Zhan; Ryan Sullivan; Weidong Cai; Wanli Ouyang; Fernando Calamante; Michael Barnett; Chenyu Wang
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

Training deep neural networks reliably requires access to large-scale datasets. However, obtaining such datasets can be challenging, especially in the context of neuroimaging analysis tasks, where cost associated with image acquisition and annotation prohibitive. To mitigate both time financial costs model development, a clear understanding amount data required train satisfactory is crucial. This paper focuses on an early stage phase learning research, prior proposes strategic framework for estimating annotated patch-based segmentation networks. includes establishment performance expectations using novel Minor Boundary Adjustment Threshold (MinBAT) method, standardizing patch selection through ROI-based Expanded Patch Selection (REPS) method. Our experiments demonstrate that tasks involving regions interest (ROIs) different sizes or shapes may yield variably acceptable Dice Similarity Coefficient (DSC) scores. By setting DSC as target, training estimated even predicted accumulates. approach could assist researchers engineers collection when defining new task based networks, ultimately contributing their efficient translation real-world applications.