Keywords:
Artificial Intelligence, CNS, Computer applications, CT, Computer Applications-Detection, diagnosis, Haemorrhage
DOI:
10.26044/ranzcr2021/R-0338
Methods and materials
BACKGROUND
Artificial intelligence and deep learning
- Deep learning (DL) is based on artificial neural networks, which train themselves on large amounts of data to perform a task.
- By automatically discovering image patterns, they can be applied to detect or classify features in images (“computer vision”).
Intracranial haemorrhage
- Intracranial haemorrhage is a time-critical neurological emergency with 5 subtypes: intracerebral (ICH), intraventricular (IVH), subarachnoid (SAH), subdural (SDH) and epidural (EDH).
- Nearly half of the mortality from ICH occurs in the first 24 hours.1
- Non-contrast computed tomography (NCCT) head scans are the practical gold-standard imaging modality of choice for detecting intracranial haemorrhage.2
Deep learning for intracranial haemorrhage detection
- Delays in intracranial haemorrhage detection translate into delays in active management, leading to potentially preventable cerebral injury, morbidity and mortality.3
- DL implementations can improve the diagnostic yield of CT, expediting detection to potentially enable earlier treatment and improve outcomes.
- However, existing DL implementations are limited by the following:
-
- Poor explainability: DL models are complex “black boxes”, often lacking transparency behind their predictions. This limits rationalisation and leads to reduced trust in these systems.
- Poor validity: Models have been tested on data derived from the same dataset they were trained on. This does not demonstrate true generalisability to new data.
- Limited clinical utility: Many models cannot subtype or locate the haemorrhage(s), limiting their clinical utility.
This study sought to implement an automated algorithm which detects intracranial haemorrhages on NCCT head studies, which addresses these limitations.
METHODS