Comparison of perfusion

Abstract

Comparison between INLINE scanner perfusion values versus values obtained from reprocessing of raw data (RECON). Several erroneous measurements exist, essentially all due to errors in segmentation, within the myocardium, or from significant blood pool contamination. I wonder if there is a way to manually contour RECON slices, perhaps with a Matlab package, so the same contours can be copied in to MBF and all BTEX variables

Internal report comparing myocardial blood flow (MBF) values from the scanner (INLINE) and from the re-processed and reconstructed data (RECON)

Stress MBF

summary(cmr_RECON$stress_MBF_global)

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.6968  1.4726  1.8170  1.8183  2.0984  3.9146 

sd(cmr_RECON$stress_MBF_global)

[1] 0.5586996

summary(cmr_inline$stress_mbf_global)

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  0.170   1.260   1.700   1.714   2.050   4.350 

sd(cmr_inline$stress_mbf_global)

[1] 0.7038344

All segments

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 -1.000   1.242   1.738   1.803   2.248   8.000 

[1] 0.8344599

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  0.100   1.140   1.610   1.721   2.200   5.180 

[1] 0.854057

Many data points are identical (on line of identity), most converge around it with good agreement. Some are errors INLINE, corrected with RECON . Hover over the plot to explore the data points.

Negative RECON values. MHCM02, 24wk, negative MBF values. S-ICD patient NOT reprocessed without T2* correction. INLINE perfusion did not segment, I contoured manually. RECON Visual maps were good, but contouring either did not compute or there was an error:

Negative RECON values, MHCM03, baseline scan. AIF from the LA of 3ch due to geometry issue. Also, S-ICD, so FLASH perfusion, without T2* correction. INLINE perfusion did not segment, I did manual contours. RECON perfusion contours not available on folder.

Very high RECON MBF values, ~6-8 with small values INLINE. MHCM20, baseline, at apex. RECON contours converge in the middle of the LV with a lot of blood pool:

MHCM05, 24w at apex, very high RECON values, also a lot of blood pooling in contours:

Rest MBF

summary(plong$MBF_RECON[plong$state == "rest"])

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.6211  0.7654  0.8106  0.9584  5.3618 

sd(plong$MBF_RECON[plong$state == "rest"])

[1] 0.2896639

summary(plong$MBF_INLINE[plong$state == "rest"])

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.1054  0.6100  0.7500  0.7680  0.9400  1.7600 

sd(plong$MBF_INLINE[plong$state == "rest"])

[1] 0.2673222

Similarly to stress MBF, several data points are identical in RECON and INLINE falling on the line of identity. Many data points converge around it with good agreement.

Low INLINE MBF, MHCM17, SICD patient, “corrected” by removing correction for apparent T2* losses.

Zero values on RECON MBF, reasonable values on INLINE, MHCM12, baseline scan at apex. SICD with FLASH perfusion with T2* correction (i.e. not reprocessed). INLINE contour errors, I did manual. RECON did not compute contours in apex. Likely because of no blood pool, cavity obliteration (despite diastolic slice):

Very high RECON MBF > 2. MHCM05, 24wk. Different RECON map, with dense focal hypoperfused areas compared to INLINE. But problem likely because of blood pool contouring.

Near zero values on RECON, between 1 to 1.5 MBF INLINE. MHCM01, 8wk, basal-mid. SICD patient, RECON without T2* correction. Wrong contour in mid slice, basal slice looks ok.

Overall impression

Most common issue is contouring. Lack of contours, or blood pool contamination. Is there a way to draw manual contours on PK’s RECON values? Perhaps via a Matlab package?

There do not seem to be biological outliers, all data should be kept.