Computer-Aided Diagnosis of Vertebral Compression Fractures Using Convolutional Neural Networks and Radiomics

Journal of Digital Imaging - Vertebral Compression Fracture (VCF) occurs when the vertebral body partially collapses under the action of compressive forces. Non-traumatic VCFs can be secondary to...     

A comparative analysis between ultrasonometry and computer-aided tomography to evaluate bone healing

An ultrasonometric and computed-tomographic study of bone healing was undertaken using a model of a transverse mid-shaft osteotomy of sheep tibiae fixed with a semi-flexible external fixator. Fourteen sheep were operated and divided into two groups of seven according to osteotomy type, either regular or by segmental resection. The animals were killed on the 90th postoperative day and the tibiae resected for the in vitro direct contact transverse and axial measurement of ultrasound propagation velocity (UV) followed by quantitative computer-aided tomography (callus density and volume) through the osteotomy site. The intact left tibiae were used for control, being examined in a symmetrical diaphyseal segment. Regular osteotomies healed with a smaller and more mature callus than resection osteotomies. Axial UV was consistently and significantly higher (p ≤ 0.01) than transverse UV and both transverse and axial UV were significantly higher for the regular than for the segmental resection osteotomy. Transverse UV did not differ significantly between the intact and operated tibiae (p=0.20 for regular osteotomy; p=0.02 for resection osteotomy), but axial UV was significantly higher for the intact tibiae. Tomographic callus density was significantly higher for the regular than for the resection osteotomy and higher than both for the intact tibiae, presenting a strong positive correlation with UV. Callus volume presented an opposite behavior, with a negative correlation with UV. We conclude that UV is at least as precise as quantitative tomography for providing information about the healing state of both regular and resection osteotomy.     

Multiplanar reconstructions of helical computed tomography in planning of atlanto-axial transarticular fixation

The objective of this study was to determine atlanto-axial bone morphometric measurements related to screw transarticular fixation technique. One hundred helical computerized tomography (helical CT) scans with volumetric acquisition, including the first and the second cervical vertebrae, were studied. The screw insertion axis according to the Magerl technique for C1–C2 transarticular fixation was the referential to select the correct oblique axial and oblique parasagittal planes obtained with multiplanar reconstruction (MPR) on helical CT. The selected measured parameters on each side of the vertebrae were C2 interarticular isthmus height and width, optimal screw length, optimal screw trajectory sagittal and axial angles, and the distance between the ideal screw trajectory and the vertebral artery groove. C2 interarticular isthmus height measured 7.75±1.27 mm, C2 interarticular isthmus width 7.94±1.72 mm, optimal screw length 39.03±2.81 mm, optimal screw trajectory sagittal angle 57.54±5.28°, optimal screw trajectory medial angle 7.90±4.05°. Isthmus narrowing under 5 mm (height and/or width) was seen in 5% of cases. In 30% of cases reconstructed parasagittal images showed the vertebral artery groove. In those cases, the distance between the vertebral artery groove and the ideal screw path was measured. This distance measured under 2.5 mm in 7% of C2 articular masses. A classification of C2 articular mass morfology was proposed. The C2 articular masses without anatomic variations predisposing to vertebral artery injury were considered type I. The C2 articular masses potentially associated with vascular injury (12%) were classified as type II. Potential risk was identified at the C2 isthmus only (3%), at the anterior portion of C2 articular mass only (7%) or at both regions (2%). According to selected criteria 18% of patients would have at least one side C2 articular mass with potential risk for the vertebral artery. In 6% of patients the potential risk was identified bilaterally. There is a great variation in the maximum and minimum values of the anatomic measurements. Therefore preoperative CT scans are very important to identify type II cases, such that the surgeon may preoperatively define the bony anatomy trough which the screws will pass.     

Fungal infection by Paracoccidioides brasiliensis mimicking bone tumor

Paracoccidioides brasiliensis infection causes a systemic mycosis originally described in Latin America but with current reports of worldwide distribution. The clinical presentation of paracoccidiodomycosis as an isolated long-bone lesion in children is quite unusual. This article describes a 10-year-old male with a lytic femoral bone lesion caused by P. brasiliensis infection that was first suspected of being of neoplasic etiology. The text also emphasizes the importance of including endemic fungal infections in the differential diagnosis of bone lesions.     

Emerging Aspects of the Body Composition,Bone Marrow Adipose Tissue and Skeletal Phenotypes in Type 1 Diabetes Mellitus

Anthropomorphic measures among type 1 diabetic patients are changing as the obesity epidemic continues. Excess fat mass may impact bone density and ultimately fracture risk. We studied the interaction between bone and adipose tissue in type 1 diabetes subjects submitted to two different clinical managements: (I) conventional insulin therapy or (II) autologous nonmyeloablative hematopoietic stem-cell transplantation (AHST). The study comprised 3 groups matched by age, gender, height and weight: control (C = 24), type 1 diabetes (T1D = 23) and type 1 diabetes treated with AHST (T1D-AHST = 9). Bone mineral density (BMD) and trabecular bone score (TBS) were assessed by dual X-ray absorptiometry (DXA). ¹H Magnetic resonance spectroscopy was used to assess bone marrow adipose tissue (BMAT) in the L3 vertebra, and abdominal magnetic resonance imaging was used to assess intrahepatic lipids (IHL), visceral (VAT) and subcutaneous adipose tissue (SAT). Individuals conventionally treated for T1D were more likely to be overweight (C = 23.8 ± 3.7; T1D = 25.3 ± 3.4; T1D-AHST = 22.5 ± 2.2 Kg/m²; p > 0.05), but there was no excessive lipid accumulation in VAT or liver. Areal BMD of the three groups were similar at all sites; lumbar spine TBS (L3) was lower in type 1 diabetes (p < 0.05). Neither SAT nor VAT had any association with bone parameters. Bone marrow adipose tissue (BMAT) lipid profiles were similar among groups. BMAT saturated lipids were associated with cholesterol, whereas unsaturated lipids had an association with IGF1. Overweight and normal weight subjects with type 1 diabetes have normal areal bone density, but lower trabecular bone scores. Adipose distribution is normal and BMAT volume is similar to controls, irrespective of clinical treatment.     

MRI assessment of bone marrow oedema in the sacroiliac joints of patients with spondyloarthritis: is the SPAIR T2w technique comparable to STIR?

Objective

To compare short tau inversion-recovery (STIR) with another fat saturation method in the assessment of sacroiliac joint inflammation.

Methods

This prospective cross-sectional study comprised 76 spondyloarthritis (SpA) patients who underwent magnetic resonance imaging of the sacroiliac joints in a 1.5-T scanner, using STIR, spectral attenuated inversion recovery (SPAIR) T2w and spectral presaturation with inversion recovery (SPIR) T1w post-contrast sequences.Two independent readers (R1 and R2) assessed the images using the Spondyloarthritis Research Consortium of Canada (SPARCC) score. We assessed agreement of the SPARCC scores for SPAIR T2w and STIR with that for T1 SPIR post-contrast (reference standard) using the St. Laurent coefficient. We evaluated each sequence using the concordance correlation coefficient (CCC).

Results

We observed a strong agreement between STIR and SPAIR T2w sequences. Lin’s CCC was 0.94 for R1 and 0.84 for R2 for STIR and 0.94 for R1 and 0.84 for R2 for SPAIR. The interobserver evaluation revealed a good CCC of 0.79 for SPAIR and 0.78 for STIR.

Conclusion

STIR technique and SPAIR T2w sequence showed high agreement in the evaluation of sacroiliac joint subchondral bone marrow oedema in patients with SpA. SPAIR T2w may be an alternative to the STIR sequence for this purpose.

Key points

? There are no studies evaluating which fat saturation technique should be used. ? SPAIR T2w may be an alternative to STIR for sacroiliac joint evaluation. ? The study will lead to changes in guidelines for spondyloarthritis.

     

Assessment of the cervical spine denticulate ligament using MRI volumetric sequence: Comparison between 1.5 Tesla and 3.0 Tesla

Background

Denticulate ligaments (DLs) are pial extensions on each side of the spinal cord, comprising about 20 to 21 pairs of fibrous structures connecting the dura mater to the spinal cord. These ligaments are significant anatomical landmarks in the surgical approach to intradural structures. To our knowledge, there is no previous study on the detection of DLs using MRI.

T1ρ and T2 Mapping of the Intervertebral Disk: Comparison of Different Methods of Segmentation

BACKGROUND AND PURPOSE:Intervertebral disk biochemical composition could be accessed in vivo by T1ρ and T2 relaxometry. We found no studies in the literature comparing different segmentation methods for data extraction using these techniques. Our aim was to compare different manual segmentation methods used to extract T1ρ and T2 relaxation times of intervertebral disks from MR imaging. Seven different methods of partial-disk segmentation techniques were compared with whole-disk segmentation as the reference standard.MATERIALS AND METHODS:Sagittal T1ρ and T2 maps were generated by using a 1.5T MR imaging scanner in 57 asymptomatic volunteers 20–40 years of age. Two hundred eighty-five lumbar disks were separated into 2 groups: nondegenerated disk (Pfirrmann I and II) and degenerated disk (Pfirrmann III and IV). In whole-disk segmentation, the disk was segmented in its entirety on all sections. Partial-disk segmentation methods included segmentation of the disk into 6, 5, 4, 3, and 1 sagittal sections. Circular ROIs positioned in the nucleus pulposus and annulus fibrosus were also used to extract T1ρ and T2, and data were compared with whole-disk segmentationRESULTS:In the nondegenerated group, segmentation of ≥5 sagittal sections showed no statistical difference with whole-disk segmentation. All the remaining partial-disk segmentation methods and circular ROIs showed different results from whole-disk segmentation (P < .001). In the degenerated disk group, all methods were statistically similar to whole-disk segmentation. All partial-segmentation methods, including circular ROIs, showed strong linear correlation with whole-disk segmentation in both the degenerated and nondegenerated disk groups.CONCLUSIONS:Manual segmentation showed strong reproducibility for T1ρ and T2 and strong linear correlation between partial- and whole-disk segmentation. Absolute T1ρ and T2 values extracted from different segmentation techniques were statistically different in disks with Pfirrmann grades I and II.

MR imaging is considered the best noninvasive method to study intervertebral disks. MR imaging allows the visualization of clearly different anatomic disk subregions, including the nucleus pulposus (NP) and the annulus fibrosus (AF).^1,2 However, routine clinical images provide a qualitative or semiquantitative assessment made by an expert.³ The need for a better understanding of physiologic and pathologic processes in the disk led to the application of quantitative techniques in MR imaging such as T1ρ and T2 mapping.^3,4For the extraction of quantitative data from a given region of interest, it is necessary to perform segmentation procedures that involve selecting the region to be analyzed.⁵ This segmentation can be manual, semiautomatic, or automatic. In studies assessing the lumbar intervertebral disk composition, different authors used different methods to perform disk segmentation with subsequent data extraction. The most common method in the literature with regard to T1ρ and T2 is to acquire small regions of interest that are anatomically based.^6–9 Authors have used standard ROIs, delineating subregions within the intervertebral disk to extract quantitative data specifically from the NP and AF.^6,7 Additional intermediate ROIs on boundaries between the nucleus and annulus have also been used.^8,9 In the latter, intermediate ROIs were implemented to compensate for the increased steps for segmentation in each image. These studies analyzed a limited number of MR imaging sections. The segmentation using a few MR imaging sections and regional ROIs allows the extraction of data more quickly than segmenting the whole intervertebral disk. Other authors chose to perform the segmentation of the disk as a whole, with the region of interest covering the NP and AF simultaneously.^10,11 The use of segmentation of only the central MR imaging sagittal section to extract quantitative data from the intervertebral disk is also very common in the literature.^12,13The intervertebral disk structure is nonuniform with differences in hydration and collagen content between NP and AF. Therefore, the extraction of different T1ρ and T2 relaxation times may be expected depending on the segmentation method used.^10,14 Despite the potential importance of using different segmentation methods in the evaluation of the intervertebral disk composition, we have not found studies comparing the accuracy, reliability, and reproducibility of the results generated by different segmentation methods. Our hypothesis is that partial segmentation of intervertebral disks, especially through standard geometric regional ROIs, will result in the extraction of different T1ρ and T2 relaxation times compared with full segmentation.