Purpose or Case Report: Hypoxic ischemic encephalopathy (HIE) is a brain injury that occurs in 1 ∼ 5/1000 term-born neonates. HIE lesion detection is a crucial step in clinical care of HIE. It could lead to a more accurate estimation of prognosis, a better understanding of neurological symptoms, and a timely prediction of response to therapy in this population. In addition, the rise of Artificial Intelligence (AI) brings hope to objectively and accurately finding HIE lesions. With public MRI data for brain tumors, Alzheimer’s Disease, and other diseases, AI has achieved significant success in MRI-based diagnosis and prognosis of these diseases. To facilitate the early prognosis and diagnosis of HIE, in this work, we focus on HIE lesion detection with MRI data using deep learning methods.
Methods & Materials: 133 patients treated in MGH between 2001 and 2018 were retrospectively collected brain MRIs. There are two modalities in the HIE lesion segmentation dataset: ADC maps and Z-score maps. We implemented the existing representative deep-learning segmentation networks to predict HIE lesions.
Results: Experimental results demonstrate that deep learning methods can be used to detect HIE lesions. However, the performance is suboptimal.
Conclusions: HIE lesion detection is a challenging task due to most data in HIE are with small diffuse lesions. A clear performance gap exists between the dice overlaps of big focal lesion segmentation and HIE small diffuse lesion segmentation using deep learning methods. More methods need to design to help HIE lesion detection to facilitate HIE prognosis and diagnosis.