MR relaxometry-based analysis of brain hemorrhages: an experimental study on a rabbit model

Magnetic Resonance Relaxometry is a quantitative MRI-based technique able to estimate tissue relaxation times T1 and T2. This approach allows increasing the MRI diagnostic accuracy mostly in case of brain neoplasia or neurodegenerative disorders in human medicine. However, few reports are available on the application of this technique in the clinical field of veterinary medicine. For this reason, in this work, we developed a relaxometry based approach on experimentally induced brain hemorrhages on rabbits. Specifically, the methodology is based on a hierarchical clustering procedure driven by the T1 relaxometry signals from a set of regions of interest selected on the T2 map. The approach is multivariate since it combines both T1 and T2 information and allows the diagnosis at the subject level by comparing “suspected” pathological regions with healthy homologous ones within the same brain. To validate the proposed technique, the scanned brains underwent histopathological analyses to estimate the performance of the proposed classifier in terms of Receiver Operator Curve analyses. The results showed that, in terms of identification of the lesion and its contours, the proposed approach resulted accurate and outperformed the standard techniques based on T1w and T2w images. Finally, since the proposed protocol in terms of the adopted scanner, sequences, and analysis tools, is suitable for the clinical practice, it can be potentially validated through large-scale multi-center clinical studies.

6 112 avoid any potential brain damage. Then the blood was manually injected into the left hemisphere of 113 each subject with a 1 ml syringe and a 25 G-1.6 cm needle perpendicularly oriented to the parietal 114 bone surface to compare the two hemispheres in the same acquisition. A) The brain "cross stitch" used as an anatomic landmark (black arrow).

118
B) The 2-mm diameter hole performed with a surgical drill 3-5 mm caudally and on the left of 119 the coronal suture.

120
C) The final location to inject the autologous blood and induce the cerebral hemorrhage.

121
122 Then the rabbit was carefully moved to the MR scanner. After the MR acquisitions, each subject has 123 been euthanized according to national guidelines, and brains were promptly fixed in 10% neutral 124 buffered formalin for histopathological investigations (see below).

126 MRI acquisition and relaxometry protocol
127 Thirteen out of sixteen subjects were included in the study since three rabbits died.
128 MR data were acquired using an Esaote Vetscan Grande scanner operating at 0.25 T equipped with a 129 Coil 4 (Esaote S.PA, Genova, Italy). The conventional data consisted of Spin Echo (SE) T1w and 130 Fast SE T2w sequences on transverse and sagittal planes, see Table 1 for the adopted parameters.   178 The basic idea is that hemorrhagic tissues will exhibit a different pattern of relaxometry as compared 179 to healthy ones. At this stage, we decided to use the original relaxometry pattern for every voxel and

278
E) The output of the hierarchical cluster overlaid on the T2 map. Voxels classified as 279 "pathological" and "healthy" parenchyma are reported in red and yellow, respectively.

280
F) The results of gross anatomy. From the photograph of the brain section, the contour of the 281 hemorrhagic lesion characterized by a linear shape that mainly involves the left thalamus (white 282 arrow) has been delineated (white dotted line).

283
G) Histopathology results. The overall brain section obtained from the microscope with the 284 hemorrhagic lesion shown in greater detail superimposed on the overall section.

285
H) The MRR classification compared to both gross anatomy and histopathology: voxels

295
E) The output of the hierarchical cluster overlaid on the T2 map. Voxels classified as 296 "pathological" and "healthy" parenchyma are reported in red and yellow, respectively.

297
F) The results of gross anatomy. From the photograph of the brain section, the contour of the

308
A) T1w transverse image of the brain; the lesion appears ill defined, hypointense, and irregular 309 in the left thalamus (white arrow).

310
B) T2w transverse image of the brain. As in T1w, the lesion in the left thalamus appears ill-311 defined, irregular but hyperintense (white arrow).

312
C-D) T1 and T2 maps, respectively. A hyperintense and irregularly shaped area is visible in both 313 maps (white arrow).

314
E) The output of the hierarchical cluster overlaid on the T2 map. Voxels classified as 315 "pathological" and "healthy" parenchyma are reported in red and yellow, respectively.

316
F) The results of gross anatomy. From the photograph of the brain section, the contour of the 317 hemorrhagic lesion that appears irregular in shape and involves the left thalamus (white arrow) 318 has been delineated (white dotted line).

319
G) Histopathology results. The overall brain section obtained from the microscope with the 320 hemorrhagic lesion shown in greater detail superimposed on the overall section.     In all the brains, a hemorrhagic lesion was detected with a pattern characterized by a 357 moderate variation of intensity from case to case (Fig 7). The injected blood caused disruption of the 16 358 surrounding neural tissue, termed mass effect (Fig 7A, B) Fragmented nuclear debris was observed 359 in damaged perihematomal tissue, in association with the presence of perihematomal edema, 360 frequently characterized by gray matter vacuolization, neuronal perinuclear halos (Fig. 7C black   361 arrow; 7D black arrow), and multifocal, mild dilation of perivascular (Virchow-Robin) spaces. Red 362 blood cells within the wall of small vessels (intramural erythrocytes) and scattered swollen 363 endothelial cells were also seen (Fig 7D white arrow). Also, multifocal cell shrinkage, with the 364 presence of scattered hypoxic-ischemic neurons was detected in almost all cases ( Fig 7D arrowhead; 365 7E arrow). For comparison, neural tissues distant from the hematoma in the same section, as well as 366 the corresponding contralateral areas in the unaffected cerebral hemisphere were examined, and no 367 significant alterations were recorded, thus allowing to exclude the occurrence of histologic artifacts 368 due to brain fixation or manipulation ( Fig 7F). 399 match only the initial part of the lesion (see Fig 6C) or half of it ( Fig 6F). Now, to quantitatively 400 validate the MRR approach, (see Section 2) the TPR and FPR for each lesion were computed. The 401 results reported in Table 2 seem promising since the mean TPR was 0.76, the mean FPR was 0.13 402 leading to an overall accuracy of 0.83.