Detection and volume estimation of artificial hematomas in the subcutaneous fatty tissue: comparison of different MR sequences at 3.0 T

Imaging techniques like computed tomography (CT) and magnetic resonance imaging (MRI) are well established in many clinical applications, and various qualitative findings and quantitative measurements derived from imaging data are used to aid radiologists and clinicians in the diagnosis and follow-up of different diseases. MRI and CT play a major role in tumor diagnosis and staging e.g. in lung cancer [1, 2, 3], breast tumors [4, 5, 6], prostate tumors [7, 8, 9, 10], and brain tumors [11, 12, 13, 14, 15], as well as in the follow-up of tumor progression [16, 17], often requiring volume measurements of tumor structures. Another specific parameter is the volume measurement of liquids in the assessment of hemorrhages. Further, volume determination has proven especially useful for measurement of liquid volumes in the assessment of hemorrhages [18], and in the diagnosis and prognosis of cerebral bleedings [19]. In the case of cerebral hemorrhage, determination of both initial hematoma volume and volume change, is essential as both factors are independent predictors of treatment outcome and mortality [20, 21]. Hence, CT and MRI protocols are well established for the determination of these critical factors [19, 22, 23, 24]. In contrast, only limited knowledge exists on the detection, volume estimation and investigation of hemorrhage in soft tissue using MRI [25], as such lesions have less therapeutic consequences.

Differently to clinical applications, in forensic medicine information on blunt force injuries, especially hematomas in the subcutaneous fatty tissue, are highly relevant [26]. Analysis of hematomas is required for forensic reconstruction and verification of a specific course of events, especially in cases involving child maltreatment or interpersonal violence. To date, the gold standard for the detection of subcutaneous hematomas due to blunt force is a detailed examination of the entire body surface including photo documentation of any visible lesions [27]. Nevertheless, subcutaneous hematomas are not necessarily visible externally, since visibility depends on many factors, such as time since impact, pigmentation of the skin or the amount of blood in the hemorrhage. In order to obtain additional information regarding blunt force associated with soft tissue injuries, the use of imaging methods, such as CT and MRI has been proven to be beneficial [28].

Due to superior contrast achievable in soft tissue and absence of ionizing radiation, MRI holds advantages over CT, particularly if used in clinical forensic cases involving victims of interpersonal violence. Although nowadays a growing interest using MRI in forensic medicine is recognized, to date there is only limited experience regarding the detection and assessment of forensically relevant soft tissue injuries [28, 29, 30, 31, 32].

Therefore, the aim of this study was the experimental determination of the detection limits of subcutaneous blood volumes in a porcine tissue model using 3.0 T MRI. Four MR sequences were evaluated regarding their potential for precise estimation of the blood volume of hemorrhages.

Blunt force injuries, such as hematomas, are a matter of special interest in forensic medicine. Although hematomas do not usually have any therapeutic consequence, their presence can have relevant implications for forensic reconstruction and verification of a specific course of events.

In order to assess the detection limit of blood volumes in subcutaneous tissue, artificial hematomas were experimentally evaluated using different MR sequences in a porcine tissue model. Firstly, the segmentability of the four studied MR sequences was evaluated, followed by examination of accuracy and precision of the volume estimation. When comparing the four sequences regarding detection of hematomas, the present study demonstrated that the TIR sequence exhibited the best segmentability rate. The most accurate results regarding volume estimation could be achieved with the T1T2w TSE sequence. In intra- and inter-observer reliability experiments the TIR and T1T2w TSE sequences showed the most reproducible results.

In general, with all four MR sequences suitable contrast properties for segmenting hematomas in the subcutaneous fatty tissue was achieved. Thus, the applicability of the selected sequences for image segmentation and volume measurement was shown. A crucial requirement for the choice of the sequences was to achieve good contrast between the lesion and the fatty tissue. The contrast achievable between different tissues is influenced by several factors, with the main factors being the relaxation behavior of each tissue. Relaxation times (T1 and T2) are intrinsic tissue properties which describe the complex process of nuclear spin magnetization returning to its equilibrium state following excitation in MRI. Additional factors relevant in achieving contrast are for example the magnetic properties (e.g. susceptibility) of tissues and saturation of specific tissue types (e.g. fat), which is dependent on the applied MR sequence.

In the gradient echo sequence (FLASH) high contrast was observed due to the differing magnetic properties of deoxygenated compared to oxygenated hemoglobin. These intrinsic tissue properties led to local magnetic field inhomogeneities, which in turn led to signal loss during the application of this imaging sequence. The applied turbo inversion recovery sequence (TIR) allowed nulling of the fat signal at an inversion time (TI) of 200 ms (3.0 T) based on the characteristically short T1 of fat. Due to the longer T1 relaxation time of venous blood, which is a major component in hematomas, the signal contribution of the lesion was not nulled enabling acquisition of images with good contrast between the hematoma and the suppressed fatty tissue. In the proton-density weighted sequence with fat saturation (PDw TSE FS) a special pulse was used to abrogate the fat signal, so that contrast between the lesion and the fatty tissue was once again maximized. In the sequence with combined T1 and T2 weighting (T1T2w TSE) the fatty tissue appeared bright due to its short T1 and rather long T2 relaxation time. In comparison, deoxygenated blood, present in hematomas, appeared darker due to its longer T1 and shorter T2 relaxation times, once again leading to good contrast in this sequence.

The contrast between the hematoma located in the muscle tissue and the actual muscle tissue itself was not adequate. This was reflected in the results of the segmentability of all lesions (intramuscular and subcutaneous fatty tissue), which demonstrated a generally lower segmentation rate. The only sequence having no location dependency was TIR, where all hematomas could be segmented independently of their location. This can be explained by the differences of the T1 relaxation times of the corresponding tissues (fatty tissue: 400 ms, muscle 900 ms, venous blood: 1500–1650 ms) [35, 36]. The other sequences did not allow a 100% segmentation rate of all hematomas but did show increased segmentability, when only hematomas in fatty tissue were taken into account.

The overall accuracy and precision of the volume estimation was poor. Averaged volumes were generally overestimated, if compared to the injected volume. If evaluating the lesions inside the fatty tissue only, there was an increase in accuracy and precision, probably due to the improved contrast between subcutaneous fatty tissue and blood. However, the volumes of lesions in the subcutaneous fatty tissue were still overestimated by all observers, especially by the radiologist. Over- and underestimation is a general issue in volume determination using imaging techniques in a medical context. Huttner et al. [37] described significant overestimation of irregularly shaped cerebral hematomas in CT images using ABC/2 technique for volume estimation. A recent study by Leddy et al. [38] noted a significant overestimation of MRI in volume estimation of breast cancer in contrast to other imaging modalities. In volume estimation of prostate cancer using different MR sequences, over- and underestimation of the actual tumor volume was observed depending on the MR sequences used, with no single sequence resulting in accurate tumor volume determination [39]. The volume overestimation observed in the present study can be explained by several factors. The inner part of the hematoma (as seen in Fig. 1) is a compact volume mainly consisting of blood, whereas in the periphery the blood intermingles with fatty tissue, dispersing around the fat lobuli and along the septa. The appearance of this outer area in MR images is dependent on the contrast and windowing settings. The chosen window parameters have a great impact on the segmentation results and therefore fixed window values were introduced in this study. Nevertheless, the individual perception of hematoma perimeters in MR images remains a main source of error in manual segmentation. Another reason for the overestimation of hematoma volumes might originate from partial volume effects, as a consequence of the limited voxel size. Especially manual segmentation by the radiologist in contrast to non-radiologist segmentation showed extended perimeters of the hematoma. Interestingly, the sequence where the least hematomas were segmented (T1T2w TSE) was the most accurate one concerning volume estimation. In the T1T2w TSE sequence fatty tissue structures were visible within some hematomas, which were not visible in the other sequences (Fig. 3), additionally leading to a better defined differentiation of the hematomas. Another effect on volume estimation could also be attributed to image resolution, which was higher for the T1T2w TSE and PDw TSE FS sequences than for the FLASH and TIR sequences. However, this effect was more prominent for the T1T2w TSE sequence.

A decreasing relative error for all observers was observed with increasing hematoma size. This can be explained by the physical model used to normalize the data. By assuming a hematoma of spherical form, the resulting error due to over-segmentation decreases with increasing hematoma radius. In this study, very small amounts of blood were examined. However, under in vivo conditions usually larger hematoma volumes would be expected, which would reduce this relative error. These factors which contributed to the volume overestimation can also be held accountable for the low intra- and inter-observer reliability obtained in this study.

One limitation of this study was that for some hematomas residual blood flowed out of the injection canal or was not completely injected leading to minor errors in the ground truth. Additionally, the acquisition protocols were optimized for achieving good contrast between blood and fatty tissue. Therefore, blood in muscle tissue was not optimally depicted.

In conclusion, this study demonstrates the potential of MRI to visualize even very small amounts of blood in subcutaneous fatty tissue. Hematomas were in general well segmentable in all of the applied sequences, with the TIR sequence showing the best segmentation rate independent of the location of the hematomas. Concerning volume estimation, the T1T2w TSE sequence was most accurate. Overall, the TIR and T1T2w TSE sequences were identified as the sequences with the highest potential for hematoma detection and volume estimation – even for very small hemorrhages.

In future, these results can be used to improve MR protocols for hematoma detection in both, forensically and clinically relevant cases. Therefore, this study provides the first step to optimize clinical forensic imaging using MRI, especially in cases where objective evidence of subcutaneous hematomas is required.