如何根据索引然后根据条件选择数据帧的行？

Question

I am running a blob detection routine on lung nodule images. For each image the blob detector returns an array of detected blobs along with their coordinates and radius. I then check if the ground truth lung nodule coordinates are within any of the blobs and return a true or false (this then becomes the training label for the next stage of the process).

The problem I am having is that for some images the nodule is within more than one blob so the image therefore has two (or more) true positive detections rather than one. For these cases I would therefore like to find the blob that is closest to the ground truth and mark that single blob as the true positive.

However I am struggling to slice the dataframe in such a way that only the positive detections per image are compared. The dataframe I have is like this:

                    Blob_Y  Blob_X     Blob_R  True_X  True_Y  Label
JPCLN001.npy 0       840.0   220.0  16.970563   817.0   346.0      0
             1       832.0   496.0  16.970563   817.0   346.0      0
             2       496.0   872.0  69.767869   805.0   483.5      1
             3       480.0   796.0  16.970563   805.0   483.5      1
             4       820.0   888.0  56.568542   817.0   346.0      0
JPCLN002.npy 5       840.0   220.0  16.970563   817.0   346.0      0
             6       832.0   496.0  16.970563   817.0   346.0      1
             7       824.0   256.0  30.169889   817.0   346.0      0
             8       824.0   172.0  16.970563   817.0   346.0      0
             9       820.0   888.0  56.568542   817.0   346.0      0

For image JPCLN001.npy I want to select the rows where label equals 1, then calculate the Pythagorean distance from (True_X, True_Y) and (Blob_X, Blob_Y) for rows 2 and 3. The blob closest to the true coordinates needs to be assigned a label of 1 and the other is assumed to be a false positive and labelled as 0.

There are four images in the dataframe that need this action performed.

I have tried doing this by selecting the relevant rows for each image and assigning them to a new dataframe, doing the distance calculation and then reinserting these rows back into the original dataframe like so:

df = blobs.loc['JPCLN061.npy']
df = df[df['Label'] == 1]

df = df.assign(dist = np.sqrt((df['Blob_X']-df['True_X'])**2 + (df['Blob_Y']-df['True_Y'])**2))
df['Label'][df['dist'] == df['dist'].max()] = 0

df.drop(['dist'], inplace = True, axis = 1)

blobs.update(df)

blobs.update(df) does not update the original dataframe (which I think is due to a mismatch between the indices of each dataframe). My method also seems rather cumbersome so if someone could help me how to do this it would be really appreciated as I've been working on this most of the day!

Answer 1

As a quick working example, how about:

df = blobs.loc['JPCLN061.npy']
df = df[df['Label'] == 1]
df = df.assign(dist = np.sqrt((df['Blob_X']-df['True_X'])**2 + (df['Blob_Y']- 
df['True_Y'])**2))
df = df.sort_values('dist', ascending=False)
blobs.loc[('JPCLN061.npy', df.index[0]), 'Label'] = 0

If you post code to create the df, I am happy to help come up with a more efficient manner!