Datasets:

Modalities:
Text
Formats:
json
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:

How to understand the localization?

#3
by dryflowerdie - opened

I understand it's axis-aligned 3D bounding boxes, but is it [x, y, z, w, h, d] or [x_min, y_min, z_min, x_max, y_max, z_max] or other type?

The question instructs the model to return values in this format
{"bounding_box": [start_slice, end_slice, x_min, x_max, y_min, y_max], "final_diagnosis": {"femur medial trochlea cartilage lesion": {"Size of lesion": (grade from 0-3), "Depth of lesion": (grade from 0-3)}}}

but i was also unable to line up the boxes properly to the images, e.g.
image
this box does not contain the femur medial trochlea cartilage. Maybe there is a preprocessing step that flips / rotates the volume before generating the bounding box values?

Rajpurkar Lab / Harvard Medical AI org

Thanks for raising this! We'll check the bounding box and update once we’ve verified what's going on.

Rajpurkar Lab / Harvard Medical AI org

The question instructs the model to return values in this format
{"bounding_box": [start_slice, end_slice, x_min, x_max, y_min, y_max], "final_diagnosis": {"femur medial trochlea cartilage lesion": {"Size of lesion": (grade from 0-3), "Depth of lesion": (grade from 0-3)}}}

but i was also unable to line up the boxes properly to the images, e.g.
image
this box does not contain the femur medial trochlea cartilage. Maybe there is a preprocessing step that flips / rotates the volume before generating the bounding box values?

Hi, could you please share the image ID for this example? Thanks!

i tried a few more examples with similar results, using the filter below, using the answer's min and max slice numbers to find and display the middle slice shown below:

SELECT image, answer
FROM train
WHERE
final_diagnosis.final_diagnosis.femur_medial_central_cartilage_lesion.size > 2
AND
final_diagnosis.final_diagnosis.femur_medial_central_cartilage_lesion.depth > 2
LIMIT 10

00m_0.C.2_9192540_20050310_10731405
00m_0.C.2_9192540_20050310_10731405
00m_0.C.2_9225592_20040526_10043703
00m_0.C.2_9225592_20040526_10043703
00m_0.C.2_9248131_20050325_10743305
00m_0.C.2_9248131_20050325_10743305
00m_0.C.2_9352883_20041026_10274706
00m_0.C.2_9352883_20041026_10274706
12m_1.C.2_9218916_20060320_10895203
12m_1.C.2_9218916_20060320_10895203
12m_1.C.2_9225592_20050526_10528505
12m_1.C.2_9225592_20050526_10528505
12m_1.C.2_9248131_20060623_11211902
12m_1.C.2_9248131_20060623_11211902
12m_1.C.2_9314101_20060316_10897609
12m_1.C.2_9314101_20060316_10897609
12m_1.C.2_9352883_20051123_10798103
12m_1.C.2_9352883_20051123_10798103
12m_1.C.2_9475286_20050504_10395812
12m_1.C.2_9475286_20050504_10395812

Sign up or log in to comment