Landmark Detection Challenge

An accurate identification of anatomical cardiac features allows more robust and accurate functional and structural analysis of the heart. Hence, anatomical landmark points are obviously essential for an automated cardiac MRI segmentation method. In this challenge, we focus on automatic landmark point detection method, which is particularly necessary for building a 3D LV model. These landmark points include mitral valve points, right ventricular (RV) insert points and base-to-apex central axis points

The Data

  • 100 cases for training set with annotation files for landmark points:
    • 2 coordinate points of mitral valve in long-axis views
    • 1 coordinate point of LV center at base
    • 1 coordinate point of LV center at apex
    • 2 coordinate points for each slice of short-axis views
  • 100 cases for validation set

Annotation Files

The cardiac landmark points are defined in a text file named as Annotations.txt. Each study has one Annotations.txt file:

PatientID: <ID>
<X> <Y> <DICOM_filename>

<X> <Y> <DICOM_filename>

[RV_INSERTS]: <npoints>
<X> <Y> <DICOM_filename>
<X> <Y> <DICOM_filename>

[MITRAL_VALVES]: <npoints>
<X> <Y> <DICOM_filename>
<X> <Y> <DICOM_filename>

Each line contains a single landmark point with the corresponding DICOM image file. All points are in image coordinates (<X>,<Y>). That means you can directly plot these points on top of the corresponding DICOM image.

If you use Matlab, you can retrieve landmark points from the Annotations.txt file by using this function: ReadAnnotationFile.m

Training set

All patients in the training set are accompanied by the landmark points annotated by an experienced analyst. You can use these landmarks for testing and training your detection algorithm

Validation set

All patients in the validation sets are also accompanied by the Annotations.txt file, but all coordinate points are marked by ‘?’ signs. Your main task in this challenge is to replace these characters with the image coordinate positions of the landmarks. You must use the same DICOM image file to detect the corresponding landmark points.

For instances, in the following example:

 ? ? DET0011201_SA3_ph0.dcm

then you must use file DET0011201_SA3_ph0.dcm to detect the base central axis point for the patient DET0011201.