The Value of Quantitative Musculoskeletal Imaging

Abstract Musculoskeletal imaging is mainly based on the subjective and qualitative analysis of imaging examinations. However, integration of quantitative assessment of imaging data could increase the value of imaging in both research and clinical practice. Some imaging modalities, such as perfusion magnetic resonance imaging (MRI), diffusion MRI, or T2 mapping, are intrinsically quantitative. But conventional morphological imaging can also be analyzed through the quantification of various parameters. The quantitative data retrieved from imaging examinations can serve as biomarkers and be used to support diagnosis, determine patient prognosis, or monitor therapy. We focus on the value, or clinical utility, of quantitative imaging in the musculoskeletal field. There is currently a trend to move from volume- to value-based payments. This review contains definitions and examines the role that quantitative imaging may play in the implementation of value-based health care. The influence of artificial intelligence on the value of quantitative musculoskeletal imaging is also discussed.

Currently, the practice of musculoskeletal imaging is mainly based on the subjective qualitative analysis of imaging examinations by radiologists. However, it is generally agreed that the integration of quantitative assessment of imaging data could increase the value of imaging, not only in research, but also for clinical practice. 1 Quantitative data can be retrieved from imaging data through various means. Some imaging modalities, such as perfusion magnetic resonance imaging (MRI), diffusion MRI, or T2 mapping, are intrinsically quantitative. But conventional morphological imaging can also be analyzed through the quantification of various parameters including size, intensity, or texture of regions or structures of interest. The quantitative data retrieved from imaging examinations can then serve as biomarkers, that is, measurable indicators of some biological state or condition. Quantitative imaging brings value to patient care because objective parameters allow for precise health decision making (e.g., to support diagnosis, determine the prognosis of a patient, or monitor therapy).
In recent years, the concept of value-based medicine has emerged as authorities are looking at ways to reduce health care costs, leading to a trend to move from payment systems based on volume to ones based on value. In this context, value is specifically defined as the ratio of the outcome over costs. 2 This review focuses on the value, or clinical utility, of quantitative imaging, taking examples from the musculoskeletal literature. In the second part, we discuss the value of musculoskeletal imaging as defined in value-based health care, that is, the ratio of the outcome over costs. We review some concepts and definitions, and then consider the role that quantitative may play in the implementation of valuebased health care. Finally, we discuss how artificial intelligence (AI) may influence the value of quantitative musculoskeletal imaging.
What Is Quantitative Imaging?
Quantitative imaging was defined by the Quantitative Imaging Biomarkers Alliance (QIBA), a working group set up by the Radiological Society of North America to standardize the use of quantitative imaging, as "the extraction of quantifiable features from medical images for the assessment of normal [findings] or the severity, degree of change, or status of a disease, injury, or chronic condition relative to normal [findings]." 3 According to QIBA, "quantitative imaging develops, standardizes and optimizes: anatomical, functional and molecular imaging acquisition protocols, data analyses; display methods; reporting structures. These features permit the validation of accurately and precisely obtained image-derived metrics with anatomically and physiologically relevant parameters, including treatment response and outcome, and the use of such metrics in research and patient care." Examples of quantifiable features range from basic diameter and volume measurements derived from structural imaging to more advanced tissue properties such as MRI relaxometry parameters and measures reflecting physiologic processes derived from functional imaging modalities. These quantitative features need to be distinguished from qualitative features (e.g., "severe vertebral fracture") but also semiqualitative features such as those used in scoring systems (e.g., Whole-Organ Magnetic Resonance Imaging Score [WORMS] or Boston Leeds Osteoarthritis Knee Score [BLOKS] scoring systems used in osteoarthritis research). 4,5 Moreover, emerging AI approaches such as radiomics and deep learning enable the derivation of novel quantitative features, also called "signatures" in this context, by mining big data using advanced statistical methods. 6,7 Clinical Utility of Quantitative Musculoskeletal Imaging Although conventional qualitative imaging, based on visual image interpretations by radiologists, already brings value to many diagnostic processes, quantitative imaging methods have the potential to exploit even better the rich information present in medical imaging data. Quantitative image analysis methods could be more sensitive to subtle abnormalities than the human eye, thus enabling earlier diagnosis. They also aid in the characterization of complex three-dimensional anatomical shapes, easily missed by visual inspection of two-dimensional slices or projections. Moreover, quantitative imaging facilitates more precise assessment of changes in longitudinal studies and comparison of patient data with reference normal values. Finally, because quantitative image analysis may often be automated, it could perform routine basis screening for multiple pathologic conditions, even on images that were originally acquired for a different clinical indication, at negligible extra cost (e.g., the example highlighted later in the Opportunistic Osteoporosis Screening section). These examples illustrate how quantitative analysis can increase the clinical utility of imaging by contributing to more accurate and earlier diagnosis, and therefore help improve patient outcomes.

Examples of the Clinical Utility of Quantitative Musculoskeletal Imaging
In this section, we provide different examples of the potential utility of quantitative imaging taken from currently available clinical applications in the musculoskeletal system.

Musculoskeletal Malignancies
Although conventional T1-weighted, fluid-sensitive sequences, and static contrast-enhanced imaging remain essential for the assessment of tumor extent and qualitative features, chemical shift imaging, diffusion-weighted imaging (DWI), perfusion imaging, and magnetic resonance spectroscopy (MRS) provide additional information regarding the physiologic and metabolic features of tumors that can be used to help categorize lesions as benign or malignant, and similarly separate viable malignant tissue from areas of posttreatment benign fibrosis. 8,9 Bone Marrow Assessment with Marrow-specific Sequences The characterization of potential bone marrow signal abnormalities often rests on conventional noncontrast spin-echo T1-weighted imaging, 10 whereby marrow-replacing lesions appear as low signal intensity lesions against a background of fatty marrow. Chemical shift imaging with in-phase and opposed-phase imaging offers a method of distinguishing true marrow replacement from signal abnormalities due to edema or hypercellular red marrow. A quantitative drop in signal on opposed-phase imaging compared with in-phase imaging of < 15% (3 T) or 20% (1.5 T) signifies the presence of a marrow-replacing lesion, whereas a drop in signal > 15 to 20% indicates an area more consistent with red marrow or edema. 11,12 ►Fig. 1 is an example of a signal abnormality at the head-neck junction of the femur that is malignant.
DWI may have added value in categorizing malignant from benign bone marrow lesions by providing threshold apparent diffusion coefficient (ADC), generally with low ADC values indicating malignancy and higher ADC values indicating a benign process 13 (►Fig. 2).
Following treatment with neoadjuvant therapy, a malignant bone tumor may demonstrate areas of posttreatment sclerosis and granulation tissue that enhance with contrast administration, thereby making the distinction of viable tumor from posttreatment sclerosis challenging. With DWI, changes in ADC values have been used to indicate treatment response (distinguishing viable tumor from posttreatment necrosis and sclerosis). 14,15 In addition, rather than performing static postcontrast imaging, a dynamic Seminars in Musculoskeletal Radiology Vol. 24 No. 4/2020 contrast-enhanced magnetic resonance (DCE-MR) sequence can be acquired that provides information regarding perfusion characteristics of a tumor. 16 In clinical practice, DCE-MR is used semiqualitatively with the construction of time intensity curves and the identification of tumor enhancement relative to adjacent arteries; early arterial enhancement in a lesion indicates malignant viable tumor, whereas a lack of early arterial enhancement suggests benign disease or treatment response. However, with DCE-MR, quantitative metrics can be derived that may play a more significant role in the assessment of musculoskeletal tumors by providing information regarding tumor heterogeneity as well as various pharmacokinetic measures. 17,18 Soft Tissue Tumors The MR imaging assessment of soft tissue tumors for malignancy relies on similar principles as those discussed for bone marrow imaging. With DWI, ADC values can be used to distinguish benign and malignant disease, [19][20][21] and they are especially useful in patients with soft tissue tumors that are indeterminate by conventional imaging. 22 DCE-MR imaging is also a useful technique for the purpose of tumor characterization, with malignant lesions demonstrating early arterial enhancement with specific enhancement patterns that benign lesions do not typically follow. 23 ►Fig. 3 shows an example of benign and malignant soft tissue tumors. Similarly, following surgery for sarcomas, DWI and DCE-MR may help in distinguishing nodular scar formation from recurrent tumor that can be challenging with conventional MRI sequences alone. 24 Finally, MRS, although a time-consuming technique, offers the ability to derive the metabolic content of an area of interest, without the use of intravenous contrast medium. Several studies showed the potential of MRS for both bone and soft tissue tumor imaging, with malignant tumors having elevated choline-containing metabolites, considered as a hallmark of cancer, above a level typically found with benign tumors. [25][26][27] An important example of an application is in the characterization of peripheral nerve sheath tumors for malignancy, whereby benign neurofibromas can be distinguished from malignant peripheral nerve sheath tumors (MPNSTs) using metabolite content, an especially useful technique for patients with neurofibromatosis type 1 who are prone to developing numerous benign peripheral nerve tumors and are at significant risk for developing MPNSTs. 28 Osteoarthritis Current imaging techniques for osteoarthritis (OA) (radiography and morphological MRI) focus on morphological damage of cartilage and other joint tissues but cannot reliably detect molecular changes in early OA, have generally limited sensitivity to OA progression, and correlate poorly with symptoms. Nevertheless, early diagnosis of OA before irreversible structural changes occur is essential because patient education, lifestyle interventions, physical and biomechanical interventions, and disease-modifying osteoarthritis drugs will be more successful and have more impact at an early disease stage. Earlier diagnosis of OA could enable an earlier start of secondary preventive measures and thus prolong the working life of the patient, benefiting both the individual and society.
Quantitative imaging techniques such as T2 mapping, delayed gadolinium-enhanced MRI of cartilage, and T1-rho have the potential to measure changes in composition of cartilage and other joint tissues occurring in initial OA stages long before the onset of morphological tissue damage, and to enable detection of OA at much earlier stages than current morphological MRI methods. 29 This will greatly improve clinical practice and clinical decisions through earlier diagnosis of OA and by identifying patients with specific OA phenotypes and risk profiles (stratification) amenable to new targeted therapies or prevention strategies (personalized medicine). Advanced machine-learning techniques for computer-aided diagnosis and prognostic modeling also show promise when applied to quantitative imaging of OA. In these methods, the computer learns from a set of examples how particular imaging features can be used as clinically relevant diagnostic/prognostic markers. Recent examples are the prediction of OA progression based on radiographs by a data-driven deep learning approach. 30,31 Opportunistic Osteoporosis Screening Osteoporosis is a highly prevalent yet underdiagnosed condition. Bone-rich data from body computed tomography (CT) scans of the chest, abdomen, and pelvis that are performed for other clinical indications are seldom used yet can provide an opportunity for quantitative osteoporosis screening.
Measurement of bone mineral density (BMD) via lumbar trabecular attenuation, particularly L1 Hounsfield unit (HU) assessment, has been studied most extensively. [32][33][34] Regionof-interest measurement of L1 trabecular attenuation is much simpler than the more labor-intensive quantitative computed tomography (►Fig. 4), yet this approach appears to be equally effective. 35 Furthermore, L1 HU correlates reasonably well with central X-ray absorptiometry (DXA), and the direct volumetric approach afforded by CT has advantages over the planar reference standard. 34 Although essentially any thoracic or lumbar vertebral body can be used, L1 is easily identified as the first non-rib-bearing vertebra, is present on all thoracic and abdominal CT scans, and correlates best with DXA. 34 From a population standpoint, L1 trabecular attenuation steadily decreases with patient age (►Fig. 5). Beyond menopause, age-matched women have a similar mean L1 HU value compared with men. 32,36 Various attenuation thresholds can be used along the receiver operating characteristics curve, providing either more sensitive or more specific assessment, but the 90 HU threshold (at 120 kV) yields the best performance for both prevalent and future fractures. 37,38 In practice, osteoporosis should generally be suspected for L1 attenuation values < 100 HU, whereas 150 HU is a useful threshold for likely osteopenia.

Muscle and Fat Quantification
Analogous to the opportunistic BMD assessment on CT scans obtained for other indications, the composition of other tissues might be analyzed at body CT. 39 In particular, quantitative assessment of both fat and muscle at abdominal CT shows considerable promise for providing additional diagnostic and prognostic value (►Figs. 6 and 7). 40,41 The obesity epidemic and the closely associated metabolic syndrome are major public health concerns throughout the developed world. 42 Although the recognized contribution of visceral (and hepatic) fat has been highlighted, differentiation of this deeper fat from subcutaneous stores is not possible through the usual clinical measures, such as body mass index. [43][44][45] CT allows for objective assessment of both visceral and subcutaneous fat that can be derived using either semiautomated or fully automated methods (►Fig. 6). 41,44 The ratio of visceral to subcutaneous fat, taken at a standardized level such as L1, may prove to be a valuable prognostic measure. 41 This may also help further our understanding of the metabolic syndrome and its close association with cardiovascular risk.
Sarcopenia is characterized by progressive and generalized loss of skeletal muscle mass, and it correlates with poor health outcomes including an increased risk for impending death. 46 CT imaging is generally considered the most accurate and reliable method to assess muscle mass and density. A variety of semiautomated and fully automated quantitative CT measures were developed within the past decade. 40,47 The L3 level at CT represents a preferred site for standardized measurement (►Fig. 7), with prior work showing that this level is optimal for assessing the psoas, paraspinal, and abdominal wall musculature. 40 As with automated BMD and fat tools, a fully automated CT-based muscle tool allows for both individualized and population-based assessment that is objective. 40 Such data can be automatically derived at abdominal CT

Osteomyelitis
MRI has demonstrated the highest diagnosis accuracy (95.6% sensitivity and specificity) for the diagnosis of osteomyelitis. 48 In the case of osteomyelitis, the presence of pus, debris, and inflammatory cells in bone marrow or even soft tissues (abscesses) may hinder the movement of water molecules. DWI probes the movement of water molecules inside tissues without the need to use gadolinium contrast-based agents (GCBAs). Thus osteomyelitis will show high signal intensity on high b values DWI. Therefore, DWI may enable evaluation, in a qualitative manner, of the presence of active infectious foci (►Fig. 8). Moreover, DWI allows quantification of the degree of restriction of water diffusion by means of the ADC values. 49 A new advanced DWI model, based on the analysis of intravoxel incoherent motion (IVIM), has been proposed to overcome some limitations of the DWI/ADC approach. This model pro-vides parameters such as f (perfusion fraction) or D Ã (pseudodiffusion coefficient) that estimate the movement of water molecules inside the capillary network, the need for GCBAs. The D (real diffusion coefficient) is also derived from the IVIM model and enables the assessment of the degree of water molecules movement without the influence of blood flow inside vessels. 50 Some authors have demonstrated that quantitative approaches may discriminate aseptic or reactive bone edema seen in Charcot foot from osteomyelitis in diabetic foot, the latter showing lower ADC values and higher transfer constants K trans and K ep values than Charcot foot 51,52 (►Fig. 9).
In the context of osteomyelitis, quantitative analysis has the potential to increase the clinical value of imaging in the following ways, some of which are still to be fully validated: -By increasing the diagnostic accuracy, especially by excluding mimics such as bone tumors and reactive (noninfectious) bone or soft tissue edema. -By providing potential cutoff values to discriminate between healthy and infected tissue as well as to differentiate among acute, subacute, or chronic osteomyelitis. -By detecting and selecting possible targets for core biopsy or fine needle aspiration at osteomyelitis based on their biological activity reducing false-negative samples and increasing the diagnostic yield of laboratory cultures.
-By providing information about the viability of bone to help determine patient outcome, possible sequels, or even to plan a hypothetical surgical intervention in the most severe areas.
-By quantifying by means of ADC and other DWI-derived parameters to evaluate patients with an absolute or relative contraindication of using GCBAs (children, pregnancy, or allergic reactions). This issue is especially useful when osteomyelitis is evaluated in diabetic patients (who usually show renal clearance problems).
-By optimizing the use of GCBAs by combining DCE-MRI with routine postcontrast imaging.
-By monitoring empirical antibiotherapy by means of parameters derived from advanced MRI sequences, allowing determination of, before morphological changes, the efficacy of antibiotherapy to adjust doses or explore other therapeutic options.

Value-based Health Care
Quantitative imaging could play a key role in the implementation of value-based health care. The concept of value-based health care relates patient health outcomes to the costs of delivering these outcomes. This idea is born out of the increased costs of delivering health care to populations, potentially making this delivery unsustainable. 2,53 These costs relate to larger populations, longer length of human life, and technological advances enabling better prevention, diagnosis, and treatment of disease. In addition, societal expectations, as well as administrative costs and perverse financial incentives to deliver more health care to individuals who may derive little or no benefit from it, are problematic.
A major challenge in the implementation of value-based health care, in particular in relation to diagnostic modalities due to their often indirect relation to patient health outcomes, is the definition and assessment of "value." Quantifying the value of a diagnostic imaging procedures is severely complicated by the nondeterministic process of visual interpretation that introduces subjectivity and intra-and interobserver variability. The value of a diagnostic procedure based on qualitative imaging therefore suffers from the same uncertainties. Quantitative imaging is a more controlled, objective, and reproducible process. It takes away some of the uncertainties in assessing the value of imaging and makes the entire diagnostic process more accountable.

Value in Health Care: What Is It?
It has become increasingly important to be able to define and measure the population and individual-level "value" of health care services if they are to be economically sustained or newly introduced. Terminology such as "cost-effective" and "valuebased" health care and "value-based payment" have been confused with one another and sometimes inappropriately applied to tests, treatments, processes, and care pathways that The plot of the mean Hounsfield unit (red dots) shows that age-related L1 trabecular bone loss is fairly linear. The normative reference ranges can serve as a quick reference while performing a CT examination for other clinical indications. Error bars indicate standard deviations, which are fairly uniform across the age spectrum. The risk for future fragility fractures increases substantially under 90 HU, and the 100 HU threshold is $ 90% specific for osteoporosis (according to a dual X-ray absorptiometry reference standard). 34 SD, standard deviation. (Reproduced with permission from Jang et al. 32 ) provide an unclear net health benefit to the patient as an individual, or to the society as a whole, to secure payment or reimbursement from insurers, governments, and patients themselves. The concepts overlap, and sometimes conflict, with one another, but they are fundamentally different with regard to the level at which they apply to the entire health care delivery system and their target. 54 Nevertheless, it is unquestionable that demonstration of the value provided by a diagnostic imaging test and the processes involved in its delivery will be key to the long-term availability of quality medical imaging services for patients and the public. This makes it imperative that value be conceptualized and measured in a manner that is both transparent and, as much as possible, independent of individual health care delivery systems and reimbursement models.
Value-based health care aims to maximize patient health outcomes achieved per unit of expenditure 54 and relates to what is of value to the individual in the context of a particular episode of care and the immediate aftermath. Inherent in this definition is the somewhat subjective nature of value, creating difficulties when this definition is applied to other patients in different situations or practice settings and when applied to longer term outcomes of the delivered care.

Increasing Value by Reducing Costs
Clearly, value, when defined as health outcomes over costs, can be increased either by reducing costs, improving outcomes, or doing both at once.
From an administrative perspective, cost reduction can be easily achieved by third-party payers, whether they be government or insurance companies, by reducing the amount they agree to pay for a test, procedure, or episode of care. Other approaches to cost reduction include bundled payments and so-called value-based payments where achievement of certain episode-specific short-term outcomes (such as lack of readmission or shortened length of hospital stay) trigger the payment. Here, value is determined by the payer, not the patient, and value-based payments are usually not aligned to longer term improved health outcomes for either the patient or society. 55 Reductions in reimbursement have been shown to be very effective in reducing overall utilization, especially in medical imaging, which is one of the major sources of all health care expenditures. 56 However, reductions in reimbursement per procedure in a fee-for-service health care system are a blunt instrument for increasing net value and have consequences related to increased out-of-pocket costs borne by patients and the public. Resulting reduced access to health care for lower income individuals, in particular, who may require health care the most because of adverse social determinants of their health, impacts life expectancy, infant mortality, emergency medical care, and productivity. These costs are borne by society as a whole. 57 The Imaging Value Chain The imaging process starts with a request by the referring physician, followed by acquisition of the images, study interpretation and reporting, and finally communication of the results. Improving the quality of provision of an imaging service improves value to the individual patient. It requires a clear focus on the steps involved in providing the service to improve appropriate utilization, reduce duplication and waste through improved efficiency, and reduce costs related to procedural errors, adverse patient outcomes, and failures to communicate critical test results. Jones et al 58 proposed a useful framework for conceptualizing these opportunities to improve the value of a medical imaging service, and Sarwar et al 59 operationalized this concept as a series of proposed "value metrics" (►Fig. 10). Before a diagnostic imaging study is undertaken, its intrinsic value to patients and society at large can be considered based on a hierarchy of efficacy.

Diagnostic Efficacy Hierarchy
The efficacy hierarchy, first conceptualized by Fryback and Thornbury, 60 is based at the lowest level on the safety of the diagnostic test, for example performance of a CT scan with the  lowest possible radiation exposure to the patient while still maintaining diagnostic image quality, and societal efficacy, such as a reduction in breast cancer deaths due to breast cancer screening, at the top. Although most diagnostic imaging tests have considerable scientific evidence supporting their safety and diagnostic accuracy, there is often much less evidence about their higher level efficacy attributes, such as changes to clinical decision making and patient outcomes. In part this is because of the difficulties in linking the outcome of a diagnostic test to a clinically relevant outcome that may be influenced by many confounders including whether an effective treatment exists.
This can create problems with illustrating the value of imaging at a payer, governmental, and societal level where critical decisions are often made (e.g., whether or not to fund a diagnostic procedure or to disinvest through cost-effectiveness analysis). Cost-effectiveness analysis is commonly used to inform population-level decisions about which procedures, pharmaceuticals, or devices will be funded or subsidized and uses a single metric to inform these decisions, the incremental cost-effectiveness ratio. Gazelle et al attempted to bridge this gap with regard to quantifying the value of diagnostic imaging test within a cost-effectiveness framework (►Fig. 11). 61

Value: The Burden of Proof
Another way to think about value in diagnostic imaging test is the so-called burden of proof. The key aspects of the diagnostic imaging test required to determine the burden of proof of its efficacy are the size of the population to whom the test or procedure may apply, its unit cost, and its anticipated clinical impact. When the population is large, the anticipated impact low, and the cost high, then societal level impact should be the level of efficacy that needs to be demonstrated, whereas when the population is small, impact high, and costs low, diagnostic accuracy may be sufficient.
The application of this concept of burden of proof to screening tests intended for a whole population, such as mammography in women aged 40 to 70, is evident when the population is risk stratified according to breast density, family history, and age, for example. In this example, screening MRI for women defined as high risk by some agreed criteria (i.e., small population, high impact, low costs) may be economically justified, or "valuable" in this context when its diagnostic performance is established without evidence of improved survival from randomized controlled trials (RCTs) of women who have and have not had breast MRI. However, if screening of all women between 40 and 70 without high pretest risk was to be considered, a higher level of efficacy (such as population-level RCT data) would be required.
How Can Quantitative Imaging Increase the Value of Imaging?
In practice, quantitative imaging has the potential to contribute to increasing the value of imaging (i.e., by improving the outcome or decreasing the cost) in different ways.

Role of Quantitative Imaging in Increasing Outcome
Quantitative imaging may improve the outcome in various ways. First, as discussed earlier, quantitative analysis of imaging may improve the outcome by increasing the accuracy. Second, the communication of actionable results (using objective data) may also improve the outcome. Finally, quantitative imaging allows the follow-up of treatment in an objective way.

Role of Quantitative Imaging in Reducing Costs
Quantitative image analysis can add value by reducing costs. First, high-resolution large three-dimensional imaging data sets can often be efficiently summarized by a few quantitative features, eliminating the need for time-consuming visual inspection by radiologists. Second, when quantitative analysis has better resilience to noise than qualitative analysis, scan time could be reduced or dose could be lowered. Third, when quantitative imaging boosts the reliability and accuracy of image-based diagnosis, it could lead to a reduction in the use of other (often more costly and invasive) diagnostic tests. Fourth, in clinical trials testing new treatments, quantitative imaging features could be used as more sensitive surrogate outcome measures or as a basis for patient selection (including only those patients who are likely to respond to the treatment), hence reducing the required follow-up and/or sample sizes and thus costs.
Costs can also be reduced by automating part of the interpretation process by using automatically generated quantitative assessment to support diagnosis. These reductions in costs should of course be balanced against the added costs of implementing quantitative imaging, due to additional acquisition time required for functional imaging sequences, for example. In many cases, however, quantification can be based on imaging data that are already acquired, using advanced postprocessing software to turn the images into quantitative measurements, thus incurring little extra cost.
Value of Quantitative Imaging in the Era of Artificial Intelligence How Can Artificial Intelligence Promote the Use of Quantitative Imaging?
One of the reasons behind the slow adoption of quantitative analysis in musculoskeletal imaging is the difficulties related to the segmentation of regions of interest, which leads to isolating tissues of interest in the image data set. When performed manually, segmentation is a time-consuming task and subject to interindividual variation. Much effort has focused on developing semiautomatic or automatic segmentation methods, but this remains a challenging task.
Machine-learning techniques have opened the doors for new automatic processes for image segmentation. These methods have been used for many years in segmentation tasks but have recently significantly gained in interest due to the rise of deep learning, which has shown impressively good performance in various computer-assisted tasks. Deep learning algorithms have been applied to various musculoskeletal applications including the segmentation and identification of bone and various soft tissues, using different imaging modalities. 7,62 However, it should be kept in mind that whether it is applied to musculoskeletal imaging or medical imaging in general, whether manual, semiautomatic, or automatic, based on machine learning or not, image segmentation remains a challenging task. 63 This is particularly true for the musculoskeletal system due to the complexity and variability of the structures of interest. Therefore, the proper validation of these methods is fundamental before their use can be generalized.
Second, AI also has the potential to promote the use of quantitative imaging by allowing the efficient communication of these data to the referring clinicians. This can be done through the automatic extraction of quantitative features and integrating of these features into standard reports. 64 The implementation of structured reporting will certainly facilitate the data-mining capabilities of AI, allowing automation of quantitative image analysis and integration of these measures into the report.

Quantitative Imaging and Radiomics
The quantitative features extracted from imaging data, such as size, signal intensity, or image texture, can also be combined with clinical, biological, or genetic data in decisionsupport models to assist physicians, either for the diagnosis and prognosis of pathologic conditions or for the prediction of therapeutic responses. 63 The process of extracting large amounts of features from radiologic images and combining them with clinical, biological, genetic, or any other type of complementary data to build decision-making models, either using conventional machine learning or deep learning, is referred to as radiomics, and it is expected to be core to the development of personalized medicine. The decision-making models built through this process may be used, for example, to diagnose a condition, stage a tumor, or predict patient survival or treatment response.

Conclusion
Quantitative imaging is crucial to bring value to patient care because objective parameters allow for higher precision medicine. Quantitative imaging should play a key role in the implementation of value-based health care. Increased use of AI and data science, allowing for further applications of quantitative imaging, will likely open possibilities to enhance the value of imaging by improving the accuracy of diagnosis, prognosis, tumor staging, and treatment response prediction.