HemorrhageNet: Automated Detection and Fine-Grained Subtype Identification of Intracranial Hemorrhage in Non-Contrast CT Imaging Using Calibrated Dual-Head EfficientNetB4

Abstract

Finding out if someone has a brain bleed, which is also called a hemorrhage is very important. We use a kind of x-ray called a cranial computed tomography or CT scan to check for this. If we do not find out about the brain bleed it can cause very bad damage to the brain or even death. In some places there are no doctors who are experts in looking at these x-ray pictures and they can get very tired from working too much. This means that sometimes people do not get the care they need . We can use computers to help look at these pictures and find out if someone has a brain bleed.

This paper is about a computer program called Hemorrha- geNet. It is a kind of program that uses pictures to decide if someone has a brain bleed and what kind of bleed it is. There are five kinds of brain bleeds: Intraventricular, Intraparenchymal, Subarachnoid, Epidural and Subdural. The program uses a kind of picture called a CT slice and it was trained on 1,200 of these pictures.

The program is very good at finding out if someone has a brain bleed. It is right 93.9 percent of the time. It never misses a bleed. The program is also good at figuring out what kind of bleed it is. It is about 80 percent of the time for each of the five kinds of bleeds.

We think that HemorrhageNet can be very helpful in hospitals. It can help doctors find out if someone has a brain bleed and what kind of bleed it is. This means that doctors can make decisions about how to take care of the person.The program uses a kind of computer model called a convolu- tional model. It was trained on 1,200 CT slices.It is very good at finding out if someone has a brain bleed and what kind of bleed it is. It can help doctors make decisions, about how to take care of people with brain bleeds.