The present study is the first attempt to estimate the metabolic pattern of the spinal canal and SC from whole-body PET/CT images. The results obtained indicate that this computational approach might be a new method for exploring the status of the SC in different conditions, besides its potential to complement the routine analysis of ALS patients.
Recognition and measurement of spinal canal and spinal cord by CT analysis
Estimation of the canal space throughout the whole spine was based on the assumption that compact bone is the human tissue with the highest X-ray attenuation coefficient and can thus be easily identified and extracted in each CT slice. This computational approach is commonly used in commercially available software for 3D representation of the skeleton, and its potential has been previously validated in our laboratory for the characterization of intraosseous volume and bone marrow metabolic activity [
15,
25,
26]. However, thresholding the Hounsfield values on the CT images is not effective in this application since the spinal canal is not directly surrounded by osseous tissue in all slices. To overcome this problem, we developed a pattern recognition method based on the Hough transform that permitted determination of canal shape also in segments between the different vertebral bodies. This approach permitted extraction of PET data and analysis of FDG uptake throughout the whole SC in a systematic fashion.
The most relevant feature of our algorithm is its fully deterministic nature with the user being asked to identify only the occipital skull border and the caudal face of D12. This plane was arbitrarily set as the caudal edge of the SC because of the limited resolution of CT images that prevented accurate evaluation of more distal segments. On the other hand, the operator-independent nature of this method virtually abolished the need for statistical analysis of its reproducibility measured in terms of either interobserver or intraobserver variability.
FDG and spinal cord involvement in ALS
As a first validation step, we applied this method to a cohort of ALS patients. This model fitted our purpose due to the pathological evidence of significant damage to the SC neurons [
1,
2]. By contrast, spinal onset ALS was associated with a slight yet significant increase in SC uptake of FDG. Several considerations support the concept that this finding reflects an increased metabolism of SC structures. First, the increase in FDG uptake was most evident in the cervical spine segments. Secondly, it was not related to abnormal regulation of serum glucose levels [
27]. Finally, the difference between controls and ALS patients virtually disappeared when the whole spinal canal was considered, most likely because of both the smoothing effect of the large volume and the spillover of radioactivity uptake into surrounding outside tissues.
The increased FDG uptake in ALS SC at least partially conflicts with the expected reduction in tissue metabolic rate caused by the neuronal loss that has been described not only in the motor cortex but also in the anterior horns of the SC in ALS patients. Similarly, it partially disagrees with the emerging prognostic value of frontal hypometabolism in subgroups of ALS patients [
4,
5,
28]. Nevertheless, it has been extensively documented in the literature that neuroinflammation is a key-signalling event in ALS [
10]. This concept originated from pathology studies showing activation of microglia and astrocytes, as well as the presence of lymphocytes and macrophages in post-mortem tissue from the motor cortex and SC of both patients and experimental models of ALS [
29,
30]. As a common interpretation, these studies suggested that activated microglia might accumulate within the degenerating areas and might contribute to propagating and sustaining the tissue damage through the release of free radicals and other neurotoxic substances such as glutamate [
30‐
32]. More recently, this mechanism has been shown to also occur in the early disease phases. In fact, different studies have shown inflammatory microglial activation in various cortical areas of ALS patients using PET imaging and different tracers targeting the translocator protein TSPO [
6‐
8]. In this line, our observation of a relative increase in SC FDG uptake might extend the previously documented pattern of motor cortex damage in ALS to lower motor neurons and might reflect inflammatory mechanisms rather than the expected consequences of motor neuron loss and subsequent SC atrophy.
FDG uptake cannot be considered per se a specific marker of microglia activation. Nevertheless, the relevance of inflammatory mechanisms on ALS progression is supported by the follow-up evaluation. Indeed, the Kaplan-Meier analysis indicated that higher FDG uptake significantly predicted a higher mortality rate. The multivariate analysis confirmed this finding and showed the independent prognostic value of SC metabolism. The observed difference in prognosis between patients with high and low FDG uptake was striking (HR 24). Nevertheless, the effective clinical potential of SC FDG uptake in outcome prediction could not be assessed in the present study because of the limited number of patients, the retrospective nature of the evaluation and particularly the fact that this hypothesis was generated by our study and needs to be validated on independent datasets.
The fact that there were 17 patients censored before death does not imply any risk of bias because these were not patients lost to follow-up, but patients with recent enrolment. Furthermore, they were similarly distributed in the two groups of patients with SUV below and above the median. A similar consideration also applies to the exclusion of patients with bulbar onset disease: this decision – justified by the focus of our computational algorithm on SC metabolism – prevents the ability to define the clinical value of this information. Finally, the relatively small number of patients studied together with the exclusion of subjects with bulbar onset ALS prevented verification of the potential added value of SC metabolism with respect to brain FDG uptake. Nevertheless, if confirmed in larger prospective studies, the prognostic significance of SC metabolic pattern would indicate a relevant role for SC inflammatory response in ALS progression.
In conclusion, the present study showed the potential of Hough transform in delineating the spinal canal and SC in clinical PET/CT scanning. As a first validation step, this method was applied to a cohort of ALS patients as a model of pathologically confirmed damage to the SC neurons. However, its use can obviously be extended to different conditions in which the possibility of extracting the spinal canal and its contents might be a useful tool to precisely evaluate the site of SC injury from whatever cause and particularly to improve the accuracy in monitoring its evolution. In this setting, the proposed computational approach to PET/CT images would permit the limitations of visual inspection to be overcome, limitations that have so far hampered the evaluation of SC damage in patients with inflammatory diseases [
33], posttraumatic conditions [
34], cancer infiltration [
35] or autoimmune/autotoxicity disorders [
36]. Similarly, its use in combination with tracers selectively targeting specific neuronal functions might contribute to the understanding of SC involvement in different diseases.
As far as ALS is concerned, the availability of this biomarker and its operator-independent nature could be invaluable for the development of new therapeutic approaches, especially in early phase clinical trials in which current entry criteria that consider only phenotype, disability severity and disease duration markedly hamper the correct identification of target patients.