Atherosclerosis is an inflammatory disease which involves several stages of plaque buildup within the arterial walls (2-4,35). The initiation of atherosclerosis is linked with the dysfunction of endothelial cells covering the walls of the blood vessels. This leads to the deposition of LDL cholesterols, within the subendothelial layers. Endothelial dysfunction and the deposition of LDL cholesterol result in the initiation of inflammatory process. This inflammatory action causes the monocytes to migrate from the lumen to the subendothelial region. Once entered into the subendothelial region, the monocytes get transformed into macrophages and absorb the lipids to become a foam cell (36,37). The accumulation of more macrophages and foam cells further forms a necrotic core that contains the lipids, foam cells, fatty streaks, and the calcified nodules (2-4,35,37). The formation of this complex plaque allows the smooth muscle cells to migrate from media layer and form a protective cover on the necrotic core to prevent its further encroachment towards the lumen (38). This whole process, when accelerated, may cause blockage in the arteries or occasionally ruptures, leading to catastrophe in the form of CVD and stroke (38). Formation of atherosclerosis is illustrated in Figure 2 (a). Since the focus of the paper is on image-based screening strategies and risk assessment, only a brief description of atherosclerosis formation is presented here and more details can be seen in the studies published by Libby et al. and Lusis et al. (3,4,38).

Figure 2

Initiation and progression of atherosclerosis in patients with diabetes mellitus. Left Figure: Diabetes accelerated atherosclerosis and Right Figure: The formation of atherosclerosis in the vessel wall (Used with permission from Cuadrado-Godia et al. (2018) (41).

Diabetes is a disease when the pancreas does not produce enough insulin or there is poor management of the insulin (39). Insulin is an essential hormone that can lower blood glucose (40). Primarily there are three categories of diabetes: (a) type 1, (b) type 2, and (c) gestational diabetes (39,40). Type 1 is characterized by poor insulin production. It is caused by an auto immune destruction of insulin producing beta cells of islets of Langerhans in the pancreases. Type 2 DM is a result of both impaired insulin secretion and resistance to its action (insulin resistance) (39,40). Gestational diabetes is a temporary disease and occurs during pregnancy (39). Women diagnosed with diabetes by standard diagnostic criteria in first trimester should be classified as having preexisting pre-gestational diabetes (40). Gestational DM is diabetes that is first diagnosed in 2^nd and 3^rd trimester of pregnancy (40). The improper management or deficiency of insulin leads to the presence of uncontrollable glucose or sugar in the blood (or hyperglycemia). High blood sugar affects the permeability of the vessel walls that carries the blood (41). The endothelium cells in the vessel wall absorb the glucose and more reactive oxygen species (ROS) are produced (41). The process of diabetes formation is shown in Figure 2 (b). Note that since the focus of the paper is on image-based screening strategies and risk assessment, the review on the pathophysiology of diabetes and of atherosclerosis has been described in detail elsewhere (6,42).

With an increased level of ROS due to hyperglycemia, there is an increase in platelet aggregation and macrophage formation, further leading to atherosclerosis (41,43). Studies have shown that atherosclerosis was more advanced in diabetes patients than non-diabetes patients, irrespective of their age (44). Further, study groups have observed the correlation of Hb1Ac (a diagnostic marker of diabetes), with subclinical atherosclerosis in Chinese (β=0.018, p=0.03) (45) and older Koreans (β=0.020, p=0.045) using multivariate linear regression analysis (46). Recently, another group (45,47,48) showed those high values of HbA1c, associated with a vulnerable plaque in the carotid artery. The association of diabetes and atherosclerosis has been an important factor in the treatment for glucose control using pharmaceutical drugs by reducing the progression of CVD (49). Additionally, Sun et al. (47) has proposed that the “determination of HbA1c and characterization of carotid atherosclerotic plaque by MR vessel wall imaging, might be useful to better select proper treatment options”. These studies suggest that monitoring diabetes systems can aid in stemming progression and vulnerability in atherosclerotic patients. Furthermore, the diabetes patients could be monitored for morphological changes in the arteries, all these factors that could lead to stroke/myocardial infarction. Thus, imaging of diabetes patient plaque burden to characterize it would be important for CV risk assessment.

Advancements in both invasive and non-invasive imaging modalities have shown the potential to accurately capture the morphological variations in the atherosclerotic plaque components (29,31,32). Screening of the carotid atherosclerosis using ultrasound is cost-effective and reliable (15). Compared to other non-invasive counterparts, Carotid Ultrasound (CUS) can provide a clear visualization of the atherosclerotic plaque, even in low-resolution images at an affordable cost (15). This was the primary reason for restricting the scope of our review to the carotid ultrasonography. It should be noted that the carotid arteries are the dominant pathways of oxygenated blood to the brain. Blockage in these arteries due to atherosclerotic plaque may lead to severe catastrophic events; such as stroke. Patients suffering from DM showed higher progression of carotid atherosclerosis, and an elevated risk of CVD/stroke events, when compared to non-diabetic patients (5). The severity of carotid atherosclerotic plaque can be investigated using the two well-known image-based screening tools: (i) carotid intima-media thickness and (ii) the carotid plaque burden (15). These two phenotypes are generally considered as the surrogate markers of the CAD that is prevalently found in patients suffering from DM. Furthermore, image-based phenotypes are closely associated with CVD/stroke events (69-81). The cIMT and the carotid plaque area are associated with a 10-15% increase in the risk of myocardial infarction (82,83). Nambi et al. (84) reported an improvement in the risk prediction of coronary heart disease, when the cIMT and carotid plaque were used as screening metrics. Stein et al. (77) recommended the use of cIMT for the screening of CHD (85). Polak et al. reported an improvement of 7.6% (P<0.0001) in the screening of CVD risk in patients using the cIMT of internal carotid artery. Lee et al. (86) presented a study that reported an association between type 2 DM and the ischemic stroke events. Furthermore, in a case where primary endpoints such as cardiovascular or cerebrovascular events are costly and require the longitudinal follow-up, cIMT can be used as an event-equivalence gold standard (87-89).

Looking at the broader usage of the CUSIP, ACC/AHA guidelines published in 2010, recommended the use of cIMT for CVD risk assessment (56). Furthermore, European Society of Hypertension (90) and National Cholesterol Education Program-Adults Treatment Panel (NCEP-ATP) III (91) also recommended the use of cIMT for the CVD/stroke screening. However, in 2012, Ruijter et al. (79) presented a study that did not recommend the use of cIMT for CVD/stroke risk screening. Another study presented by the same authors also made the similar recommendations which did not support the use of cIMT for routine screening of CVD in DM patients (80). These studies ignited the debate on the use of carotid ultrasound image phenotypes for the CVD/stroke risk assessment. This further resulted in modifying the ACC/AHA guidelines, which did not agree with their earlier statement (56). As a result, they did not recommend the use of cIMT for the routine CVD risk screening (19). The primary issue with the studies presented by Ruijter et al. (79,80) was the measurement of cIMT within the 10 mm plaque free-region of the common carotid artery. Further, the authors did not measure the plaque in the carotid bulb and the internal carotid artery (92). In general, atherosclerotic plaques are uncommon in CCA segment and are more aggressive in the carotid bifurcation i.e., at origins of the internal and external carotid arteries. This may be one of the reasons why a little improvement was observed in CVD risk when cIMT was added along with FRS (79,80).

Although, both cIMT and CP are commonly considered for the preventive screening of the carotid atherosclerosis, their accurate link with CVD is dependent on how the measurement method was conducted. At present, the measurement of these CUS image-based phenotypes (CUSIP) can be followed using three types of methods: (i) manual, (ii) semi-automated, and (iii) automated measurement.

Manual measurement of cIMT involves the delineation of lumen-intima (LI) and media-adventitia (MA) boundaries (interfaces), by the physician, using a limited number of manually placed sample points. Although manual measurements are universally available and inexpensive, they are prone to the inter- and intra-operator variability and may not generate the reproducible results at different time instants for the same patients (80,87).

Semi-automated measurement techniques involve the use of edge delineation software packages that define the limit of cIMT in a pre-specified carotid segment selected by the operator (87,93). Such software packages facilitate the cIMT measurement at different carotid segments, with sufficiently large number of measurement points (80,87,93,94).

Recently, automated measurements have also been used to measure the cIMT and carotid plaque area (95-100). Such techniques automatically delineate the LI and MA borders. There is no need to manually place the region-of-interest. The tracing of carotid plaque is also possible using such automated packages throughout the length of ultrasound scan (95-100). The number of points (or vertices) are interpolated into a set of fixed number, equi-distanced points (typically normalized to 100), and B-spline fitted to smooth the boundaries. These interfaces (or boundaries) are then used to compute the bidirectional distance between the two boundaries (95-100) using polyline distance algorithm that provides the mean cIMT value and is considered as the final cIMT value for the carotid artery segment (87,93).

Current manual and semi-automated techniques restrict the cIMT measurement within a 10 mm plaque free region of the common carotid artery (69-81). However, it should be noted that the morphological variations in the LI and MA interfaces can extend outside the 10 mm region of the common carotid artery. Thus, there is a possibility to miss the elevation of the atherosclerotic plaque near the bulb region that may not capture from the plaque, in the CUS image, by the semi-automated software. In such cases the measurement of LI and MA throughout the length of ultrasound scan, so-called a full-length measurement, may prove to be beneficial (95-100). A recent group by Suri et al. has proposed several such full-length automated measurements of CUSIP (41,95-102). Furthermore, in order to assess the 10-year risk of patients, the similar group has also proposed the automated measurement of 10-year CUSIP that combines the effect of conventional risk factors and the baseline automated CUSIP (13,67). This integrated approach has been discussed in detailed in next section of this review.

In a study of 496 patients, Ikeda et al. (103,104) reported a significant association between cIMT and ABI and showed the usefulness of this association in predicting the CAD. Further, the authors also evaluated the correlation between automated full-length cIMT and the SYNTAX score. They finally concluded that both automated CUSIP such as CIMTave and cIMTV were significantly correlated with the SYNTAX score. Figure 4 (a) shows the significant correlation between SYNTAX score and average cIMT (r=0.225, P=0.00001) and Figure 4 (b) indicates the significant correlation between SYNTAX score and cIMTV (r=0.230, P=0.00001). Authors also reported the strong association between automated full-length phenotypes and ABI. Ikeda et al. (105) used computerized methods for bulb edge detection and estimated IMT in the 10 mm segment near the bulb. A similar team had developed a fully automated system for IMT detection and achieved a precision of 92.67% (105). The same team has observed an improvement in the correlation of auto-IMT and SYNTAX score when compared with the sonographer IMT (Sono-IMT) (r=0.4675 vs. 0.323, p<0.0001) (Figure 5) (104). AtheroEdge^TM offers an automated tool for assessment of image-based phenotypes (IMTV, IMT_avg, IMT_max, IMT_min and mTPA) that can be used for risk assessment (106). Table 2 shows the studies that validated the automated CUSIP for screening of the CVD risk.

Figure 4

A. Correlation of cIMT with SYNTAX score (r=0.225, P=0.00001); B. Correlation of IMTV with SYNTAX score (r=0.230, P=0.00001) (Used with permission from Ikeda et al. (106)).

Figure 5

A. Correlation of Auto-IMT with SYNTAX score. B. Correlation of Sono-IMT with SYNTAX score. (Used with Permission from Ikeda et al.(104)).

A recent clinical trial has used the two image-based screening tools such as coronary computed tomography angiography, and MRI for detection of unrecognized myocardial scar in subclinical coronary atherosclerosis and this indicated severe stenosis in patients suffering from type 2 DM (108). Out of the 348 subjects enrolled, 322 were classified as patients without occult myocardial scar and patients with at least one occult myocardial scar using MRI. The mean (SD) age of the study population was 60.18 years (6.48). This is summarized in Table 3. This clinical trial shows that at least 61 (20.4%) patients had significant stenosis with 6 (2.0%) having 100% occlusion in patients without an occult myocardial scar. In patients with at least one occult myocardial scar, 12 (52.2%) patients had significant stenosis and 1 (4.3%) had 100% occlusion.

Lacroix et al. (109) argued that due to the prevalence of atherosclerosis in diabetes, there was a high chance of >60% stenosis, especially in men with CKD or those having an ABI<0.85. Thus, screening and 10-year risk assessment would help both in the prevention of CVDs and for better disease management. It was observed during a follow-up study by Kao et al. (110), that the association with atherosclerosis was more pronounced (0.80 vs 0.64 mm, p <0.01) in patients who had CVE. Another study by Elias-smale et al. (111) analyzed that the addition of cIMT information to the FRS improved the reclassification of risk. Screening in this study helped in better risk classification. Further Polak et al. (77) discussed that ICA-IMT improved the risk stratification in the Framingham Offspring Study. These further assert that screening information along with traditional risk scores would help in better identifying intermediate and high-risk patients. Table 4 summarizes the clinical trials studied in this review.

Thus, the use of preventive screening in assessing atherosclerosis was emphasized as this could save both life and money at a modest cost for screening. Draman et al. (112) reported the case of a 62-year old man with diabetes who came for an annual clinical review. Screening for regular risk factors did not reveal stenosis and CAD. The patient’s ECG (113) revealed ST-elevation myocardial infarction (STEMI) and coronary angiogram showed stenosis and coronary artery disease. Treatment at the right time saved his life but the patient subsequently passed away in one-year. Reason of death was not specified but the presence of severe stenosis cannot be ruled out. Two more similar case reports are summarized in Table 5. Diabetes patients have a higher chance of stenosis development and this may lead to MI, stroke, other cardiac events and sometimes sudden death (114). Duration of diabetes also increases the chance of cardiovascular events (115). Screening and risk assessment are crucial especially in the presence of diabetes and traditional risk factors.

The assessment of CVD risk is vital in identifying the intensity of risk prevention. Ultrasound screening is essential in assessing the plaque burden. The morphology, geometry, and stability of plaques are significant in cardiovascular risk assessment (115). The geometry assessments involve measurement of image-based lumen diameter (LD) (116-118), IMT (96,119-122) and inter-adventitial diameter (IAD) (116-118,123). Saba et al. observed that the IAD was strongly associated with plaque score (a measure of plaque burden) than LD (123). Recently, it was analyzed that cloud-based (AtheroCloud^TM) automated LD measurement was accurate and had better inter- as well as intra- operator reproducibility (124). Figure 6 (a) shows the ROC and area under the curve (AUC) of both AtheroCloud^TM and sonographer (manual) IMT markings. The AUC of the cloud-based system was higher compared to the sonographer markings (0.99 vs 0.81) and had better accuracy when compared with the ground truth (sonographer readings) (124). This has been represented by the ROC curve shown in Figure 9 (175).

Figure 9

ROC and AUC for AtheroCloudTM and sonographer IMT markings. (Used with permission from Saba et al. (175)).

Other than blood biomarkers, image-based phenotypes, plaque echolucency, the severity of stenosis, and texture-based features are frequently used for cardiovascular risk stratification. The association of cIMT with cardiovascular risk was noted earlier and it was observed that an increase of 0.1 mm in cIMT was associated with a 25% increase in cardiovascular risk (125). Further, changes in the plaque composition lead to the formation of plaque that is vulnerable ad prone to rupture (126,127). The plaque echolucency also changes as a result of the change in composition. Moreover, it is noted that cardiovascular risk is higher in subjects with increased plaque echolucency (128).

Research studies have confirmed that diabetes patients have more echogenic and calcified plaques compared to non-diabetes subjects (129). Another study group illustrated those patients with hypoechoic plaque and stenosis >50% have a higher chance of stroke (130). Further, Pedro et al. (131) used plaque echo-morphological features such as plaque ulceration, echolucency and echo-heterogeneity obtained from B-mode ultrasound along with stenosis severity for assessing neurological risk. The study observed 77% accuracy in predicting symptomatic plaques (131). Araki et al. used plaque morphology evaluated by machine learning algorithms for risk stratification and the team achieved a classification accuracy of 98.43% (132-135).

Plaque characterization is vital in risk stratification and prediction of CVE (136). Plaque characterization involves classification of symptomatic and asymptomatic plaque. Texture-based features are frequently used for such classifications (137-139). A research study observed that the heterogeneous plaques were symptomatic when compared with homogeneous plaques. They analyzed that the heterogeneous plaques were linked to lipids and intra-plaque hemorrhages (115). Furthermore, several manual and semi-automated algorithms are available for plaque characterization (117,140,141). Suri et al. (142) developed Atheromatic^TM system, similar to Symptosis™ (143), a computer-aided diagnosis system, for highly accurate plaque characterization and the team observed 83.7% accurate results (144-146). It used DWT, radon transform and higher-order spectra along with the textural features. Authors observed an accuracy improvement to 90.66% when the degree of stenosis was also used for the classification (147). The similar group further used texture-based features and traces transform and obtained an accuracy of 93.1% (147). Further learning-based methods (a class of AtheroEdge^TM system) have shown success in plaque characterization and achieved better performance compared to conventional methods (148,149). Recently, another study demonstrated an 18% improvement in risk stratification using integrated machine learning (ML) approach over the conventional ML method. The integrated ML approach used 34 image-based phenotypes and 13 conventional risk factors for the risk stratification (68).

In 2017, the global healthcare expenditure on the patients suffering from diabetes was $850 billion (8). In India, an average expense-per-year for a patient suffering from DM is $119.4 (7), which is ~10% of the average per capita income (176). There are 72 million (7,29,46,400) people affected by diabetes in India and ~1 million people (9,97,803) died in 2017 due to DM (7). Around 51% (4,22,10,300) (7) patients with diabetes are undiagnosed and hence does not get proper treatment. Only ~30 million (total 72 million – undiagnosed 42 million) undergo treatment for DM (7). The average annual per capita income in India is ~$1778 (176) and an annual burden of diabetes ambulatory care per person is $207.25(177). Thus, a total annual burden for 72 million people affected by diabetes is $15,118.66 (~$207.25 x 72 million). The complications due to CVD add further cost burden. The overall indirect cost per person in diabetes is $182.22 (178) and thus, a total indirect cost for all the patients with diabetes in India is 13.1 billion ($182.22 x 72 million). Mean age of onset of diabetes is 45.4 (±10.9) (179) and ~50% diabetics die at age <60 years (7). This is far less than life expectancy in India (180). The economics of diabetes is summarized in Table 6. The cost involved in diabetes treatment from various studies is shown Table 7. Recenlty, Kang et al. (181) presented a study that showed the total life-years lost due to DM and other CVD complications. The total life-years lost by the Asian population are: (i) 6.10 years due to diabetes alone, (ii) 16.68 years due to diabetes and stroke, (iii) 21.10 years due to diabetes and myocardial infarction (MI), and (iv) 23.80 years due to diabetes, stroke and MI. If the patients are screened using non-invasive methods, complications due to DM can be minimized, mortality rate can be reduced, and quality of life can be improved. If 23.8 years in life are saved as a result of screening and follow-up, then $42, 316.4 ($1778 x 23.8) can be saved for the family and the total saving for 72 million diabetes patients will be ~$3046 billion ($42, 316.4 x 72 million). This might be easily achieved by a modest cost spent on screening for atherosclerosis. Screening using non-invasive methods such as CUS imaging can be used to prevent any undiagnosed casualties such as silent CVD or stroke events which are common in patients with diabetes. Kotabagi et al. (182) reported five cases of MI related post-mortem study in which survival time after the appearance of first symptom was just 2 hours to 5.5 hours. In another study, 219 patients out of 728 in the study group died in the first five year of the onset of symptoms and 81.9% of deaths were due to ischemic heart disease (183). Screening for atherosclerosis and allied complications followed by appropriate therapies and lifestyle changes may improve the quality of life. Mostly, the rural health is affected due to unavailability of expert medical care and the heavy financial burden caused by the complications. Improvement through awareness programs and expert consultations in rural areas through telemedicine can reduce the financial burden (184). Furthermore, at a modest cost of screening for atherosclerosis, improvement in quality of life and better management of complications can be achieved (185). However, all these screening paradigms need to be tested and proved in prospective randomized trials prior to widespread implementation.

Diabetes and atherosclerosis have serious complications. Studies have analyzed that diabetes has several manifestations such as oral (150), musculoskeletal (151), rheumatological (152-154) and cutaneous (169). Furthermore, nephropathy in diabetes was a manifestation of insulin resistance and genetic factors (155). Studies inferred that heart failure was another manifestation of diabetes (156). It was observed that the integrated 10-year risk computed using AECRS1.0_10yr and HbA1c are associated with the prevalence of cardiovascular risk. Various studies have indicated that the higher values of IMT (125), TPA (41,157) and HbA1c are linked with high-risk in CVD/stroke or CAD. Furthermore, plaque echolucency was associated with cardiovascular risk (128), and >50% stenosis (130) was associated with stroke risk. Researchers observe that HbA1c levels >6-8% (diabetic) and >5-6% (non-diabetic) are at a higher risk. We strongly believe that preventive screening would benefit intermediate and high-risk patients. Furthermore, it was observed that interventions to reduce the deviation of biomarkers and HbA1c would benefit high-risk patients. Certain medications for treating diabetes were observed to be effective in treatment of coronary atherosclerosis (49). Glucose lowering therapies using sodium-glucose linked transporter (SGLT-1) inhibitors have shown to lower CVD risk in diabetics (158). New medications such as empagliflozin and liraglutide reduce mortality in patients with diabetes and CVD (159,160). Additionally, it is observed that SGLT2 inhibitors such as empagliflozin and canagliflozin reduce hospitalization due to heart failure (161,162). Further, early interventions using lipid-lowering drugs such as statins and proprotein convertase subtilisin kexin type 9 (PCSK9) have been shown to reduce cardiovascular risk in secondary prevention and show promise for primary prevention as well (163,164). Treatments other than pharmacological drugs involve lifestyle changes and exercise (165). Increased physical activity, plant-based diet, completes quitting of tobacco and proper glycemic control results in an improved disease management. Risk factor management involves nutrition therapy, obesity management, blood glucose control, lipid control and blood pressure (BP) control. Pharmaceutical treatment involves aspirin therapy, glucose control medicines, BP and lipid-lowering medicines (166). Lifestyle remains a core means of reducing CV risk. Studies have observed that subjects who do regular exercise have good glycemic control (167). Further, HbA1c levels are controlled in subjects who do aerobic exercise (167). Moreover, the rapid decrease in sugar levels in subjects who do exercises was reported by García-García et al.(168). Fenkci et al. observed a decrease in insulin resistance in subjects doing aerobic exercise (169). Another study observed that yoga reduced the CVD risk, and this study was validated using the FRS and 10-year CVD risk score (170).