Menggunakan Arsitektur Hybrid Cascade EfficientNet-V2 dan ConvNeXt
Clinical Decision Support System untuk deteksi dini dan triase perdarahan intrakranial dari CT scan kepala
Try Detection →⚠️ For academic research and demonstration purposes only
Intracranial Hemorrhage (ICH) is a critical medical emergency with mortality rates reaching up to 40% within 30 days if not diagnosed and treated promptly. Early detection is crucial for improving patient outcomes.
The radiologist ratio in Indonesia, especially in Sumatra, is significantly below WHO recommendation of 1 radiologist per 20,000 population. This shortage increases the risk of delayed diagnosis and missed critical findings.
Manual CT scan interpretation is subject to inter-observer variability and can be time-consuming, especially in emergency situations where rapid decision-making is essential for patient survival.
This system functions as a Clinical Decision Support System (CDSS) designed for early triage assistance. It aims to support radiologists and emergency physicians by providing rapid preliminary analysis. This system does not replace radiologists but serves as an assistive tool.
Intracranial hemorrhage (ICH) is bleeding within the skull, a critical medical emergency requiring rapid diagnosis and treatment. ICH can occur in different locations within the brain, each with distinct clinical implications.
Bleeding within brain tissue, often caused by hypertension or trauma
Bleeding into the brain's ventricles, affecting cerebrospinal fluid circulation
Bleeding in the space between the brain and surrounding membrane
Blood collection between the brain and its outermost covering
Bleeding between the skull and the outer membrane of the brain
This system uses advanced deep learning techniques to analyze brain CT scans and detect multiple types of intracranial hemorrhages simultaneously using a hybrid cascade architecture.
DICOM files are processed using Hounsfield Unit (HU) windowing with three specialized windows (Blood, Brain, Bone) to create optimal input for the neural network, enhancing hemorrhage visibility across different tissue types.
A cascade architecture combining EfficientNet-V2 for fine-grained local feature extraction and ConvNeXt for global spatial context understanding. This hybrid approach enables comprehensive multi-label classification across all ICH subtypes.
The system employs three independent trained models running in parallel to provide robust predictions with cross-validation. Results are presented comparatively to increase diagnostic confidence and reduce false negatives.
Institut Teknologi Sumatra
Universiti Malaysia Perlis
This project represents a collaborative academic research initiative between Institut Teknologi Sumatra (Indonesia) and Universiti Malaysia Perlis (Malaysia), focusing on the application of artificial intelligence in medical imaging and clinical decision support systems.