Page 85 - Contributed Paper Session (CPS) - Volume 4

CPS2131 Philip L.H. Yu et al.
networks with huge amount of data and parameters, which have
revolutionized the artificial intelligence in achieving the outstanding results on
many challenging tasks. The medical imaging community has taken notice of
these pivotal developments. The deep learning has been widely applied to
medical image analysis, demonstrating the state-of-the-art performances on
many medical image analysis tasks, including classification, detection and
segmentation.

2. Methodology
The Hong Kong Hospital Authority’s Clinical Management System (CMS)
has well-established records of all patients admitted to the public hospitals for
all types of acute ischemic stroke in 2016. The study population was stratified
using disproportionate random sampling methods, and the patients’ selection
criterion is announced in another published paper (Tsang et al., 2019). Total
150 patients were sampled from the ischemic stroke database in CMS. The
data includes both CT images and some on-set clinical information, e.g.: side
of weakness. The samples were then randomly split into 120 for model training
and 30 for validation.

The MCA ground truth was independently evaluated by two
cerebrovascular disease specialists. Any discrepancies were resolved by
consensus. The segment labels were manually drawn through software FSL

74 | I S I W S C 2 0 1 9