Brain MRI Pulse Sequences Classification

About Project

This is official code of our package brainmri_ps. We provide a machine learning based tool to automatically classify Brain MRI series into different pulse sequence types:

• FLAIR
• T1C
• T2
• ADC
• DWI
• TOF
• OTHER

Installation

Install via pip:

pip install brainmri_ps

Usage

Load pretrained models:

from brainmri_ps import PulseSequenceClassifier
classifier = PulseSequenceClassifier("mobilenet_v2").from_pretrained()

Name	Input Resolution	#Params (M)	MACs (G)	Test Accuracy	Pretrained
MobileNet V2	256	2.23	0.42	100.0	✓

Example - predict from a study:

In  : classifier.predict_study("*/1.2.840.113619.6.388.6361536015762131135133837693432843617")
Out :
{
    "1.2.840.113619.2.5.1821162425615901145251590114525252000":   "ADC", 
    "1.2.840.113619.2.388.57473.14165493.12954.1590103413.819":    "T2", 
    "1.2.840.113619.2.388.57473.14165493.12954.1590103413.822":   "DWI", 
    "1.2.840.113619.2.388.57473.14165493.12954.1590103413.823":   "T1C", 
    "1.2.840.113619.2.388.57473.14165493.12954.1590103413.821": "FLAIR"
}

Function predict_study does the following steps:

• Read all dicom files in a study folder and group them into series by SeriesInstanceUID field
• Determine the orientation plane (axial, sagittal, coronal) of the series by using the ImageOrientationPatient field
• Predict and return the pulse sequence types of axial series (ignore the non-axial ones)

Contact

Issues should be raised directly in the repository. For further support please email us at:

• [email protected]
• [email protected]