Weakly Supervised Deep Learning-based Intracranial Hemorrhage Localization

• DOI: 10.5220/0010825000003123
• Publication Date: 2022-03-03T11:01:08Z
• Authors: Jakub Nemcek; Tomas Vicar; Roman Jakubicek
• Keywords: Sørensen–Dice coefficient; Dice; Position (finance); Supervised Learning

Intracranial hemorrhage is a life-threatening disease, which requires fast medical intervention.Owing to the duration of data annotation, head CT images are usually available only with slice-level labeling.However, information about exact position could be beneficial for radiologist.This paper presents fully automated weakly supervised method precise localization in axial slices using position-free labels.An algorithm based on multiple instance learning introduced that generates likelihood maps given slice and even finds coordinates bleeding.Two different publicly datasets used train test proposed method.The Dice coefficient, sensitivity positive predictive value 58.08 %, 54.72 % 61.88 respectively, achieved from dataset.