The proposed custom-made, 2D perfusion DSA algorithm was safe and achieved the quantification of infrapopliteal angioplasty for the treatment of CLI, with a technical success rate of 72%.
A MTT decrease after angioplasty was noted in four out of five cases, indicating an increase of tissue perfusion following successful revascularization. This was reflected in the clinical outcomes, as wound healing was achieved in four out of five cases at 6-month follow-up. Moreover, pre-procedural and post-procedural variation of MTT was quantified, and the percentage of increase or decrease of foot perfusion following endovascular revascularization was calculated. Notably, an increase in tissue perfusion was noted in all ROIs directly supplied by the revascularized arteries. However, in patient #1, in which pedal arch occlusion revascularization was not technically feasible, an increased tissue perfusion was detected only in the area supplied by the revascularized anterior tibial artery (Fig. 6a, ROI #1). Moreover, in this case, the perfusion calculated in ROIs supplied by the posterior circulation at the level of the heel was decreased after angioplasty of the anterior tibial artery (Fig. 6a, ROIs #2, #3, and #4). The authors speculate that this phenomenon could be explained either by a technical error owned to a mismatch between pre-procedural and post-procedural ROIs attributed to unperceivable limb movement or by the fact that angiosome-targeted revascularization of the occluded anterior tibial artery and the absence of communication between the anterior and posterior circulation led to redistribution of foot perfusion favoring the recently revascularized anterior tibial artery area. On the other hand, in patient #2, in involving both the anterior and the posterior circulation (Fig. 6b). This is in line with previous studies indicating the quality of the pedal arch significantly influences wound healing rather than angiosome-directed angioplasty [16].
Finally, in patient #5, no arch was visible after revascularization and an increase in perfusion was noted only in restricted tissue areas around the revascularized anterior tibial artery. This patient underwent major above the knee amputation 1 month after the procedure, further highlighting the fact that the achievement of vessel patency is not synonymous to clinical success, as well as the impact of pedal arch outflow for limb salvage, as previously reported [17].
In this study, CBF, CBV, and MTT parameters were extracted from TIC curves and evaluated for patients’ condition improvement having CLI. These parameters were computed by means of a custom-made FCM-based segmentation algorithm that manages to detect only vessel information towards performance improvement and processing time decrement. The proposed segmentation algorithm omits non-valid information from vessel background and could be independently used to each commercial software.
As regards available software packets, differently from our study, most approaches employ parameters from TICs such as “area under the curve” and “peak concentration.” The aforementioned computed parameters were also employed by Galanakis et al. [18] in a relatively small sample of patients. Both results from the two studies are in agreement, demonstrating clinical importance and feasibility.
In addition, the properties of the proposed segmentation algorithm provide the employment of small ROIs on the vessel mask for analysis differently from other software packets that process information regarding lower foot area derived from large ROIs [13, 14, 18,19,20]. Other studies in the literature also tried to quantify DSA perfusion angiography on patients with diabetes using quantification parameters extracted by TICs using appropriate software implementations [13, 14, 19, 20]. These works have shown that quantification techniques using the aid of specialized software can monitor successfully patients’ condition and provide a useful tool for clinicians. Although these studies’ approach is similar to that here proposed, a direct comparison cannot be made due to the small sample enrolment, absence of a “reference standard” for segmentation, and different quantification parameters.
The proposed method is a semiautomatic reader-dependent method which needs some experience of use. As regards the registration procedure, it is mandatory only when substantial differences between DSA examinations are present. To avoid this processing time-expensive additional step, specific foot casts are under construction to standardize and ensure reproducibility of the pre-DSA and post-DSA imaging procedure. When these differences between pre-DSA and post-DSA image sets are omitted, the algorithm can be used in a real-time analysis. Regarding the current dataset and the radiologist’s experience, a small amount of training was necessary since the majority of ROIs’ selection was successful without the need of repetition.
This feasibility study has limitations. The most important one is that the number of patients investigated was small. Therefore, regardless of the aforementioned promising results, the small number of limbs included in the analysis deteriorates the generalization properties of the proposed approach. Another limitation is that visual estimation and not quantitative vessel analysis was used in order to assess procedural success. Nevertheless, procedural success was a secondary endpoint of this study, while both visual estimation and quantitative vessel analysis can only assess patency of the treated vessel and not the hemodynamic parameters of foot perfusion that were quantified by the proposed angiography perfusion software. As regards the vessel mask extraction procedure, it is dependent on the user-specified ROIs leading to initialization values fed to the FCM algorithm. The FCM variant used in this study is a pixel-based clustering method that suffers from noise and intensity variations that may categorize pixel areas or objects to the same cluster when the desired outcome is different. For some problems such as the construction of the vessel mask in this study, a preinitialization of the clusters’ centers. is required. ROIs taken on fully perfused vessels with high intensity values may lead to under-segmentation of small vessels that are not fully perfused and therefore reach lower intensity values. As there was no reference standard used regarding vessel extraction, there is no indication regarding the performance of the segmentation algorithm process. Regarding the TIC extraction and MTT calculation process, it is dependent on the total frame number and frames per second rate. Although a polynomial fit is applied to the TICs, small differences between estimated and real values should be expected. The registration process is also dependent on user-defined points that indicate the same anatomical areas on pre-therapy and post-therapy vessel masks. Since the mean intensity values of PBV, MTT, and PBF of the chosen ROIs were calculated, relatively large enough ROIs should be selected to minimize calculation errors due to unmatched areas. It should be mentioned though that the vessel extraction process is implemented for lesser computational times and ease of pre-therapy and post-therapy image evaluation process, assuming that the aforementioned limitations do not affect diagnosis. Limb immobilization was essential for image acquisition and post-processing, as movement artifacts do not allow correct post-processing of the acquired images. For this reason, perfusion imaging was not feasible in two patients as adequate immobilization using a soft strap could not be achieved. Perhaps the use of more efficient immobilization methods could expand the applicability of the method even in less cooperative patients, while this method would be easy to perform in cases that have been selected to undergo revascularization under general anesthesia.
Due to the relevant technological improvement of noninvasive imaging modalities, currently recommended pre-procedural and post-procedural imaging for patients undergoing peripheral endovascular interventions includes duplex ultrasound, computed tomography angiography, and magnetic resonance angiography. Intra-arterial DSA is reserved only for intra-procedural imaging and in cases of unresolved radiological issues with noninvasive imaging. However, the proposed algorithm could be used in general clinical applications and other noninvasive modalities such as magnetic resonance techniques, where a similar protocol as that used in this study could be utilized by the proposed algorithm for quantification purposes.
The proposed algorithm produced the exact same results after the 10 times repetition. This is attributed to the vessel and background ROIs’ selection for FCM initialization that leads to the same local maxima detection and finally the calculation of the same cluster centers. As regards the intraobserver and interobserver variability, the ICC calculation showed excellent agreement for the intraobserver variability and good agreement for the interobserver variability for each parameter (CBF, CBV, and MTT). This shows that the proposed method may be used as an objective alternative to visual inspection.
In conclusion, quantification of infrapopliteal angioplasty outcomes using this newly proposed, custom-made, intra-procedural PDSA algorithm was performed using PBV, MTT, and PBF maps. Limb immobilization remains a challenge especially in non-cooperative patients. Further studies are deemed necessary to determine its clinical role in peripheral endovascular procedures.