A picture is worth a thousand words: a history of diagnostic imaging for lymphoma

Abstract

The journey from early drawings of Thomas Hodgkin’s patients to deep learning with radiomics in lymphoma has taken nearly 200 years, and in many ways, it parallels the journey of medicine. By tracing the history of imaging in clinical lymphoma practice, we can better understand the motivations for current imaging practices. The earliest imaging modalities of the 2D era each had varied, site-dependent sensitivity, and the improved accuracy of imaging studies allowed new diagnostic and therapeutic techniques. First, we review the initial imaging technologies that were applied to understand lymphoma spread and achieve practical guidance for the earliest lymphoma treatments. Next, in the 3D era, we describe how anatomical imaging advances replaced and complemented conventional modalities. Afterward, we discuss how the PET era scans were used to understand response of tumors to treatment and risk stratification. Finally, we discuss the emergence of radiomics as a promising area of research in personalized medicine. We are now able to identify involved lymph nodes and body sites both before and after treatment to offer patients improved treatment outcomes. As imaging methods continue to improve sensitivity, we will be able to use personalized medicine approaches to give targeted and highly focused therapies at even earlier time points, and ideally, we can obtain long-term disease control and cures for lymphomas.

In 1666, Marcello Malpighi described “a disease of lymph nodes and spleen that was uniformly fatal,” resulting in the first documentation of lymphoma. Nearly 200 years later, in 1828, Dr Robert Carswell exhibited a collection of drawings and paintings of his patients, including one which caught the attention of his friend, Dr Thomas Hodgkin (Figure 1). Hodgkin noted, “…I was struck with one representing a greatly enlarged spleen, loaded with large tubercles of a rounded figure and light colour. I immediately recognized [it]…”² In 1832, Hodgkin assembled seven case reports, including Carswell’s patient, and published “On Some Morbid Appearances of the Absorbant Glands and the Spleen” offering the first formal description of the pathologic characteristics of lymphoma. These initial descriptions and illustrations motivated Samuel Wilks in 1865 to examine the pathologic specimens and christen the clinical diagnosis as Hodgkin’s disease (now Hodgkin’s Lymphoma [HL]).³ While these early descriptions and drawings are far less detailed than modern imaging, they were similarly intended to convey information about the disease and illuminate the patient’s broader clinical picture. Thus, even in its earliest forms, imaging can help clinicians to identify a disease, to target therapies, and to assess whether this disease was adequately treated. By tracing the history of imaging in clinical lymphoma practice (Figure 2), we can better understand the motivations for current imaging practices.

Figure 1.

Earliest known illustration of Hodgkin’s Lymphoma. A painting by Carswell as part of five plates in watercolor to show pathologic features. They were first published in 1898.¹

Figure 2.

Timeline of Important Dates. Significant events in lymphoma diagnosis, imaging, and staging are shown.

Optimal imaging of lymphoma requires overcoming several diagnostic challenges. First, although lymphoma is principally a malignancy of the lymphatic tissues, most often involving lymph nodes and spleen, it may involve any organ and may involve single or multiple sites. Second, the general term lymphoma includes separate entities HL and non-Hodgkin’s lymphoma (NHL), and with regard to the latter, multiple differently classified subtypes have varied clinical behavior and prognoses. Third, while there are numerous diagnostic options – there is no perfect modality – requiring judgment as to which approach is best for a particular scenario. More specifically, each imaging technique has inherent advantages and limitations, whether related to technology (e.g. resolution), side-effects/complications (e.g. radiation exposure), efficacy (e.g. sensitivity, specificity), expertise required to interpret the images, availability (e.g. geographic), convenience (e.g. time/space required), and expense. All these considerations have influenced the imaging of lymphoma from the past to present and will continue to shape the future.

The 2D imaging era

The earliest imaging modalities each had varied sensitivity that was often site-dependent, and the overall increasing accuracy of imaging studies allowed for novel treatment approaches, although lymphoma management was often constrained by the practical limits of available imaging.⁴ For example, after Scholtz described radiotherapy for lymphoma in 1902,⁵ Rene Gilbert, another pioneer in radiotherapy, treated HL patients with low-dose radiotherapy in the 1920s.⁶ This practice may have been premature, but it highlights both growing confidence in and reliance on imaging modalities to lead treatment innovation. In the following section, we describe initial imaging technologies that were applied to understand lymphoma spread and achieve practical guidance for the earliest lymphoma treatments.

Imaging modalities

X-ray

At the turn of the twentieth century, only 6 months after the discovery of X-ray (XR) by Willhelm Roentgen, physicians were already using radiographs to identify bullets and broken bones during the Balkan War. To advocate for clinical use of XR, or roentgenograms, Marie Curie even drove portable XR equipment onto battlefields in France during World War I. The clinical utility of the earliest XR machines was limited by long scan times and high radiation exposures, but by 1910, diagnostic and therapeutic XRs were already in clinical use.⁷

At diagnosis, roughly 60% of all HL patients have mediastinal involvement, and roughly 20% of early-stage patients present with bulky mediastinal disease.⁸ Chest X-ray (CXR) was used to evaluate the chest for evidence of lymphoma, response to treatment, and any associated abnormality (e.g., infection, or complications from radiation or chemotherapy). The ratio of mediastinal mass diameter to a reference transverse chest diameter on CXR was used to identify “bulky” disease, which altered management (Figure 3). The German Hodgkin Study Group (GHSG) and the European Organization for Research and Treatment of Cancer (EORTC) both included mediastinal bulk in staging because of its prognostic value. Although XR is largely replaced by modern imaging for staging and follow-up, CXR remains an initial and inexpensive means of evaluation for thoracic symptoms in an undiagnosed lymphoma patient.

Figure 3.

Chest X-ray of classical Hodgkin’s lymphoma. 31-year-old female presenting with (a) bulky mediastinal disease and stage IIAX classical Hodgkin’s with (b) good response seen after ABVD for six cycles and involved field radiotherapy.

Venocavography/Pyelography/Barium Enema/Upper GI and Small Bowel Series

By injection or ingestion of contrast agents in conjunction with XR, clinicians could identify intrinsic involvement by lymphoma in thoracoabdominal blood vessels with inferior venocavography, with kidneys with i.v. pyelography (IVP), and the GI system. These techniques could also provide indirect evidence of abdominal adenopathy related to space-occupying displacement of organs by enlarged nodes. For lymphoma, these techniques were mostly replaced by more sensitive modalities throughout the 1980s and 1990s.

Lymphography

XR in combination with lymphography, also known as lymphangiography (LAG), involving injection of a contrast media to opacify the lymph vessels, allowed evaluation of lymphatic channel abnormalities. Lymphography became the diagnostic gold standard for diagnosing HL, with lower extremity lymphography identifying previously undetectable abdominal lymphomas (Figure 4).

Figure 4.

Example of lymphography. 42-year-old female with nodular sclerosing Hodgkin’s lymphoma with lymphography showing left para-aortic lymph nodes larger than the right (white arrow) concerning for early involvement.

Lymphography had numerous complications and contraindications, leading to Dr John Ultmann describing it in 1966 as “...probably the most useful as well as the most abused [imaging modality].”⁹ The injection of lipid dye was associated with numerous complications including hypersensitivity and fat embolism, and contraindications include both cardiovascular disease and pulmonary disease. While it was an excellent technique for direct imaging of the size and internal architecture of retroperitoneal nodes, it required skill and expertise to perform LAG and interpret the imaging. Thus, misinterpretation by less-experienced users was a major limitation, as well as the absence of visualization of mesenteric nodes.

Conventional linear tomography

Before computers could reconstruct imaging sections through CT in the 1970s, a source was moved with a detector to provide cross-sectional visualization within a focal plane. This conventional linear tomography (CLT) was mainly employed for whole lung tomography to evaluate possible pulmonary involvement, but it was also sometimes useful to evaluate the mediastinum and hila for questionable adenopathy detected on CXR. For mediastinal HL, a suspected tumor was recommended to have whole lung CLT, and a finding in one plane would trigger reimaging in the perpendicular plane. Although there were concerns about radiation exposure and false-positive lung nodules, throughout the 1980s CT (discussed below) eventually replaced CLT.

Ultrasound

When CT availability was limited in Europe and other parts of the world, ultrasound (US) was commonly employed. Although appealing due to the absence of radiation, availability, and relatively low cost, early bi-stable sonography images were of low resolution and very operator-dependent. Subsequent real-time US provided more consistent and higher resolution images, but images could be limited by body habitus and bowel gas shadowing, reducing usefulness. When abdominal nodes could be demonstrated, size was the only criteria for evaluation identifying adenopathy larger than 2 cm. The performance of US ranged and appeared to have site specificity, but the technique could only give evidence about node size, not malignant involvement. For periaortic and retroperitoneal LNs, US had a 93.8 and 87.5% true-positive detection rate, respectively, and US was an improvement over XR.¹⁰ However, sonographic imaging of the spleen allowed accurate determination of spleen size, but discrete splenic nodules were not always sufficiently different in sonographic “texture” to be detected.

Scintigraphy

The invention of the rectilinear scanner in the 1950s allowed automated measurement of the scintillation of ionizing radiation released from radioactive tracers, but they had unclear utility in lymphoma. By the late 1960s, the use of the γ camera allowed improved sensitivity.

In 1967, Gallium-67 scintigraphy (GaS) was originally employed as a bone-seeking agent but was noted to have uptake in some lymphomas.

Initially, it did not have wide clinical appeal, attributed to long scan times producing complicated images requiring specialized training to read as well as the isotope’s unfavorable long half life.¹¹ The resolution of GaS was low and small volume disease sites escaped detection. Additionally, activity in a residual mass post-treatment implied residual disease, but false-positives and -negatives were problematic. However, by the 1990s, GaS was routine for post-treatment evaluation in lymphoma.¹² When GaS was compared to inferior venocavography, GaS had comparable sensitivity (31% vs 42%), true-positive rate (92% vs 83%), and true-negative rate (47% vs 59%), demonstrating that GaS could potentially replace more invasive imaging approaches like venocavography and lymphography.¹³ An example of GaS is shown in Figure 5.

Figure 5.

Example of Gallium-67 Scintigraphy. 40-year-old male with nodular sclerosing Hodgkin’s lymphoma who presented with a 2 × 0.4 cm right submandibular lymphoma node (white arrow) with increased radiogallium.

Lymphoma staging and diagnosis

In a landmark paper in 1950, Vera Peters identified HL staging by extent and location,¹⁴ later including symptoms.¹⁵ In 1965, Kaplan modified the staging to specify side of the diaphragm to aid in radiation design and include a Stage IV (bone, pulmonary, and gastrointestinal lesions) reflecting confidence in disease detection.¹⁶ According to the 1970 Report of the Committee on Hodgkin’s Disease Staging Procedures chaired by Saul Rosenberg, staging of HL should include a chest roentgenogram, IVP, lymphogram, and skeletal system XR or CLT.¹⁷ Additionally, conditional whole-lung-CLT could clarify suspicious CXRs, and venocavography could clarify equivocal LAG or IVP. In the 1960s, Stanford physicians arguing that HL spread continguously introduced the staging laparotomy with splenectomy, liver biopsy, and LN sampling, but this technique was ultimately abandoned because it carried a risk of mortality, the procedure itself was identified as a risk factor for disease relapse and secondary malignancies, and NHL patterns of spread were unique.¹⁸ Although staging laparotomy was not the standard of care, it provided an important gold standard to verify radiographic detection of infradiaphragmatic disease. The sensitivity and specificity of various imaging modalities are shown in Table 1.

Table 1.

Diagnostic performance of various imaging modalities for HL, NHL, and overall lymphomas

	HL	NHL	Overall
	Diagnosis	Diagnosis	Diagnosis
X-ray	Sn:39–77% Sp:92–100%
Lymphography	Sn:85–95% Sp:98%	Sn:49–91%	Sn:85–100% Sp:75–93%
US	Sn:60–97% Sp:90–100%%		Sn:30–64% Sp:96%
Ga-67 Scintigraphy	Sn:64–95% Sp:90–98%	Sn:69–90% Sp:90%	Sn:67–85% Sp:87–100%
CT	Sn:65–96% Sp:41–92%	Sn:86% Sp:67%	Sn:40%
MR	Sn:60–100% Sp:94–100%		Sn:35–100% Sp:31%
PET/CT	Sn:86–94% Sp:87–96%	Sn:89% Sp:100%	Sn:94%

Sn, Sensitivity; Sp, Specificity.

Although US had limited use in staging, ultrasound was often employed as an initial diagnostic modality for abdominopelvic lymphomas. A comparison of staging by US to venocavography, GaS, and lymphogram in primarily HL showed US could detect abdominal disease with sensitivity of 64% and a specificity of 96%.¹⁹ An early study of US for retroperitoneal LNs concluded that, “…echographic study of lymphoma patients will outline almost all enlarged LN groups in the abdominal area and will assess some areas not seen by conventional roentgen examination. In those patients considered for curative radiation therapy, both lymphangrography and echography are desirable in addition to other roentgen studies.”²⁰

Scintigrams could clarify bone abnormalities, osseous pain, and elevated alkaline phosphatase, but they were not diagnostic of hepatosplenic disease.²¹ An early pioneer of Ga-67 scintigraphy stated that although it was of, “...no value in disease staging... [scintigraphy] has a unique role in the various aspects of treatment evaluation in lymphoma.” Demonstrating this concept, a retrospective study of HL patients with GaS showed overall 72% correct classification based on surgical staging.²² Ga-67 uptake in viable tumor but not fibrosis led to GaS detecting recurrence, discussed below.

Although CT imaging has largely replaced LAG, LAG could clarify a negative CT as it is more sensitive for architectural abnormalities in normal-appearing LNs. The concordance of LAG and CT for involvement of retroperitoneal adenopathy is 90%.²³ Lymphoma patients with bone marrow biopsies (BMB) in conjunction with lymphography had lower relapse rates compared to staging by other modalities.²⁴

Prognosis and response assessment

In a report from David Karnofksy from Memorial Hospital in 1966, the only obligatory follow-up exam for HL was a 3-month follow-up XR.²⁵ In 1973, one early trial for advanced HL, employed XR of chest and positive retroperitoneal nodes before each chemotherapy cycle, with complete remission (CR) only after XRs resolved.²⁶ By the early 1980s, a trial testing chemotherapies in advanced HL required CXR, bone, liver, and spleen scans, and abdominal lymphography after every two chemotherapy cycles.²⁷

Although Ga-67 baseline uptake in lymphoma was not specific for disease involvement, there was a correlation between positivity and residual disease and CR on GaS had improved patient prognosis.²⁸ Recommendations were for residual Ga-67 uptake to warrant further non-cross-resistant treatment,²⁹ with at least 3-week wash-out. While scintigraphy could identify active disease, 95% of initially unsuspected relapses were missed.³⁰

In HL and aggressive NHL patients, positive GaS after one cycle of chemotherapy had worse outcomes.³¹ Ga-67 uptake in NHL after 4–6 cycles of chemotherapy predicted residual disease with poorer patient outcomes.³² Patients with advanced stage, aggressive NHL had a CT chest/abdomen/pelvis (CAP) and GaS, and negative early restaging GaS was predictive of progression-free survival.³³

The 3D imaging era

In the early 1970s, the introduction of CT and MR allowed better visualization of patient anatomy fundamentally altering the approach to imaging in lymphoma; however, unlike functional imaging approaches, neither technology could adequately resolve whether residual masses contained metabolically active lymphoma. In the following section, we describe how anatomical imaging advances replaced and complemented conventional modalities.

Imaging modalities

CT

With the rapid development and availability of CT, diagnosis of and determination of extent of disease dramatically improved, and one imaging modality allowed simultaneous evaluation of multiple anatomic sites and organ systems with ever-increasing resolution, albeit with the costs of radiation exposure and higher expense. The advent of spiral CT and multi-detector row CT in the 1990s allowed for thinner slices and faster image acquisition, which minimized breathing artifacts. CT could identify mesenteric, retrosternal, axillary and paraaortic LN enlargement³⁴; however, abdominal disease on CT was frequently pathologically negative.³⁵ Although CT images displayed anatomic abnormalities, pathology remained presumptive and required confirmation. On CT, spleen size and discrete nodules for biopsy could be readily determined, although these suspected areas of disease are sometimes isodense with adjacent normal tissue on contrast imaging. Follow-up CT examinations could demonstrate response to treatment, but some cases left residual masses, so-called presumed “sterile” masses. In treatment naïve mediastinal HL, when compared with clinical and histological staging, GaS, CT, and MR showed a sensitivity of 90, 96, and 100%, respectively, with scintigraphy and CT together having 100% sensitivity.³⁶ Of note, restaging CT scan specificity for local progression was only 45% vs >90% for other modalities as CT is less able to resolve residual disease vs fibrosis,³⁷ and the PPV for both CT and MR was 11 and 33%, respectively, compared to 67% for GaS.³⁶

MR

After the rapid advancement of CT, MRI became the next area of imaging evolution. Based upon different physical properties, abnormal tissue could be now demonstrated more reliably in splenic nodules and in bone marrow. However, the ability of MR to detect involvement of nodes and other organs remained site-specific. Although they do not expose patients to radiation, MRI scanners are expensive, and the many possible sequences are time-consuming making evaluation of multiple body parts impractical for most patients, although whole-body MRI can be beneficial in some clinical scenarios like reducing radiation exposure for young patients or follow-up of lymphomas without PET uptake. Historically, MRI was most useful for the detection of infradiaphragmatic lesions like involvement of the spleen.³⁸ Also, MRI could aid in identifying bone marrow involvement detecting patients with previous false-negative bone marrow biopsies³⁹ and was highly concordant with scintigraphy.⁴⁰

The homogeneity of NHL on MR was correlated with low-grade disease, and the more homogenous high-grade lesions have improved survival.⁴¹ MR signal intensity patterns can indicate active disease even in the case of stable or decreased tumor size; however, within 6 months of treatment, MR may detect necrosis and inflammation.⁴² Although MR detects residual mediastinal disease in HL,⁴³ ultimately, MR could not discriminate non-malignant LN pathologies.⁴⁴ Efforts to make MRI more common in lymphoma evaluation are being made through combination with PET imaging, additional sequences, and MR spectroscopy, but currently it is best utilized for specific or targeted examinations (e.g. brain/spinal cord, and bone).

SPECT

Single photon emission computed tomography (SPECT) imaging is a functional imaging technology that is similar to positron emission tomography (PET) imaging, but it measures emitted γ rays instead of positrons resulting in less spatial resolution.⁴⁵ The invention of a hybrid scanner performing both CT and SPECT functions, the SPECT/CT, improved image co-registration and removed motion artifacts allowing more clinical utility of SPECT by improving anatomical localization.⁴⁶ Planar and SPECT/CT GaS before lymphoma treatment had sensitivities of roughly 80% and specificities approaching 100% in detecting disease compared to clinical staging with CT and US, while post-treatment sensitivity for residual disease was 92%.⁴⁷

Lymphoma staging and diagnosis

In 1989, the Cotswolds Meeting was held in England to discuss issues surrounding evaluation of HL patients citing that CT and MR were in routine clinical use. The committee recommended staging with CXR, CT CAP at 1 cm intervals, and bipedal lymphography, although both CT and lymphography were not needed, with confirmation of abnormal findings by scintigraphy, US, and MRI.⁴⁸

Initially, investigators were relatively cautious resulting in recommendations to only use CT in conjunction with conventional diagnostics methods, although it was acknowledged that abdominal CT was likely superior to radionuclide scanning and CT was likely body site and comorbidity-specific. CT was used to diagnose abdominal disease, enlarged retroperitoneal LNs, and anterior mediastinal disease.⁴⁹ In NHL, abdominal CT staging could alter staging in 14% of patients and detect residual disease in 43%.⁵⁰ Compared to CXR, CTs can detect additional disease in a site-dependent manner in 0–15% of cases and alter treatment in 9.4%.⁵¹ CT, lymphography, and BMB were compared with clinical exam, showing that the three modalities upstaged NHL patients.⁵² CT and lymphography are discordant in 4–33% of cases, favoring abnormal lymphographic findings with normal CT.⁵³ However, lymphangiography was shown to underestimate the volume and extent of periaortic nodal masses, and CT could detect lymphography-missed renal, mesenteric, and spleen disease.⁵⁴

For the earlier GHSG trials, patients generally had staging procedures involving XR, CT CAP, BMB, serum chemistry, while abdominal US, bone scans, staging laparotomy with splenectomy, and liver biopsy were only performed depending on when the trial was initiated and reflecting the underlying expected burden of disease.

Lymphoma response assessment

According to the 1999 NCI International Working Group for criteria for response assessment in lymphoma, “CT scans remain[ed] the standard for evaluation of nodal disease. Thoracic, abdominal, and pelvic CT scans are recommended even if those areas were not initially involved because of the unpredictable pattern of recurrence in NHL. Studies should be performed no later than 2 months after treatment has been completed to assess response.”⁵⁵ An example of CT response assessment is shown in Figure 6.

Figure 6.

Example of CT response. The patient referenced in Figure 4 had a baseline CT scan (a) and received ABVD for six cycles with an enlarged level IB cervical node (white arrow). End of treatment CT (b) showed residual adenopathy with a positive biopsy. He received radiation for refractory disease, with a restaging scan (c) showing good response.

Trials in the 1990s began to use CT for adaptive therapy. For example, GSHG HD6, comparing COPP/ABVD to COPP/ABV/IMEP followed by radiotherapy for bulk or residual tumor, imaging response was assessed following the fourth and eighth cycles of chemotherapy and following radiotherapy, with at least partial remission (PR) defined as disease >2 cm on CT after cycle eight not receiving additional treatment.⁵⁶ In the GHSG HD8 trial, testing CMT with extended field radiotherapy or involved field radiotherapy, response XRs were obtained during treatments, but restaging via CT was planned 2 weeks following chemotherapy and 8 weeks following radiotherapy and every 3 months thereafter.⁵⁷ The GHSG HD9 study, which tested different chemotherapies with radiation for initially bulky or residual disease, assessed response after the fourth and eighth chemotherapy cycles using CT, sonography, scintigraphy, bone marrow biopsy, or liver biopsy depending on the sites of initial disease. Residual tumor on CT received 40 Gy.⁵⁷

The PET era

The introduction of automated, total body scanning allowed better localization of radioisotopes and by the mid-1990s, PET scans were already being used to understand the response of tumors to chemotherapy.⁵⁸ By the 2010 publication of the GHSG HD10 and HD 11 trials, important trials testing modern chemotherapy and radiotherapy regimens, the authors acknowledged that PET may be useful in risk stratification during and after chemotherapy in future studies.^59,60 The following section summarizes the work that established the importance of PET imaging in lymphoma.

Imaging modalities

FDG scintigraphy

Although scintigraphy was already used by the late 1990s, it was primarily for skeletal surveys. Labeled 2-F-18-Fluoro-2-Deoxy-D-Glucose (FDG) glucose analog can indicate hypermetabolic states suggestive of malignancy. FDG lacks a hydroxyl group that prevents it’s metabolization after uptake by cells, so increased detection of the F18 radioisotope corresponds to increased glucose uptake.⁶¹ In cancer, tumors have increased glycolysis and glucose consumption.⁶² FDG uptake measured as a standardized uptake value (SUV) identifies malignancies, provides biologic insight, and locates metastases.⁶³ A converted gamma-camera system for FDG scintigraphy had similar diagnostic performance with GaS for lymphoma grade and response.⁶⁴

Positron emission tomography

In the 1970s, PET was developed⁶⁵ to measure cerebral glucose metabolism⁶⁶ but it was not used routinely in clinical medicine until 2001.⁶⁷ PET could quantify tumor perfusion, evaluate tumor metabolism, and trace radiolabeled cytostatics,⁶⁸ and discriminate viable tumor.⁶⁹ In 2000, a combined PET/CT scanner provided precise anatomical co-registration and short scan time resulting in fewer false-positives and false-negatives relative to PET alone.⁷⁰

The advent of PET scanning represented a significant advance in the evaluation of lymphoma, particularly when the technology allowed combined PET and CT or MRI images. Combined anatomic and physiologic information has improved initial staging and subsequent response to treatment. Multicenter trials for lymphoma showed that PET had an 8–43% higher accuracy for disease detection compared to other modalities.⁷¹ Because of more extensive PET-identified disease, between 20 and 40% of cases had a change in treatment plan.⁷² The aggressiveness of primarily NHLs was shown to be correlated with PET SUV^73,74 with an SUV >10 having an 81% specificity for aggressive lymphoma,⁷⁵ and uptake was increased to a greater degree than Ga-67.⁷⁶ Residual activity within a “sterile” mass strongly indicates residual disease and provides a target for histologic sampling if needed.

While PET/CT has become a mainstay in the diagnosis and treatment of lymphoma, the FDG utilized only indicates metabolic activity and is not specific for lymphoma. Low-grade NHL may not be detectable, and areas of activity may relate to infection or other neoplasms. Identification of lymphoma-specific agents for PET/CT would be a major advancement, but this is yet to be a reality.

Lymphoma staging and diagnosis

In 2014, recognizing the importance of PET imaging, the International Conference on Malignant Lymphoma International Working Group (ICML IWG) recommended that PET/CT with contrast-enhanced CT be used for staging most nodal lymphomas,⁷⁷ with CXR no longer required.⁷⁸ PET resulted in improved lymphoma staging, additional detected lesions, and improved cost-effectiveness^79,80 compared to gallium^81,82 and CT^83–85

In HL patients, FDG-PET and conventional staging both detected 96% of involved LNs, but PET was more sensitive and discriminated false-positives on CT resulting in altered treatment strategies.⁸⁶ PET and conventional methods were discordant 20% of the time mostly involving previously missed, FDG-positive sites,⁸⁷ like extranodal disease.^88,89 Even in indolent NHL, PET downstaged 30% of patients and upstaged 16% resulting in altered management in 34%.^90,91

Bone marrow involvement occurs between in approximately 50% of NHLs and up to 15% of HLs, and initially PET-detection showed approximately 80% agreement with BMB.^92,93 However, depending on the lymphoma subtype PET and BMB concordance was as low as 57% with a high false-positive rate for HLs and a high false-negative rate observed for NHLs, and, ultimately, BMB cannot be omitted.⁹⁴

Lymphoma response assessment

By 2007, the need to revise the response assessment to account for PET scans was noted.⁹⁵ In 2009, a joint workshop between nuclear medicine physicians and hematologists in Deauville, France, met with the primary goal of reaching a consensus on criteria for interim-PET interpretation.⁹⁶ Among general recommendations regarding PET in lymphoma, they outlined criteria for scoring PET staging and treatment response that remains widely used (Table 2). The 2011 ICML IWG recommendations were for mid-treatment imaging PET-CT being superior to CT alone to assess early response, but no change in treatment should be based solely on PET.⁵⁵ The EOT PET was recommended as standard of care for remission assessment in FDG-avid lymphomas with biopsy recommended in the case of residual disease, and the possibility remaining for additional therapy based on PET-positivity.⁷⁷ When PET was evaluated relative to ICML IWG criteria in DLBCL patients, PET was shown to increase the number of CR patients and completely eliminate the prior equivocal categories.^55,97

Table 2.

Deauville criteria for PET staging and response⁷⁸

Score	Criteria
1	No uptake above background
2	Uptake ≤ mediastinum
3	Uptake > mediastinum but≤liver
4	Uptake moderately >liver
5	Uptake markedly higher than liver and/or new lesions
X	New areas of uptake unlikely to be related to lymphoma

Imaging-independent prognostic assessments for HL and NHL were improved by PET.^98–101 PET/CT is also more accurate for restaging lymphoma relative to PET or CT alone.^102,103 Changes in interim PET scans have been noted as soon as 7 days after initiating treatment.¹⁰⁴

The significance of interim PET scans in HL was established through numerous studies varying cycles of chemotherapy showing negative PET-1,¹⁰⁵ PET-2^106–108 PET after multiple cycles,^109,110 and EOT PET^111–115 were all associated with improved outcomes. This included site-specific studies,¹¹⁶ pre-transplantation PET,^117,118 and residual masses.^119–122 An example of PET-based response is shown in Figure 7.

Figure 7.

Example of PET response. 68-year-old male with stag IIA non-bulky, unfavorable classical HL, with pre-treatment PET scan (a) showing left cervical disease (white arrow) and complete response on PET-2 after ABVD (b).

The sensitivities and specificities for PET detection in lymphoma are shown in Table 1. Trials using PET-directed therapy for lymphoma treatment are summarized in Table 3.

Table 3.

Trials using interim PET to adapt therapy

Study	Patients	Prior to PET	Interim PET	Decision based on PET	Outcome
H10	Early-stage HL, favorable and unfavorable	two cycles of ABVD	PET-2	PET-negative: ABVDx2 (favorable) and ABVDx4 (unfavorable) PET-positive: Escalated chemoradiotherapy	RT could not be omitted for PET-2 negative
RAPID	Early-stage HL, 2/3 favorable and 1/3 unfavorable	3-cycles ABVD	PET-3	PET-negative: omitted radiotherapy	RT could not be omitted PET-3 negative
HD16	Early-stage HL, favorable	two cycles ABVD	PET-2	PET-negative: omitted radiotherapy	RT could not be omitted for PET-2 negative
HD17	Early-stage HL, unfavorable	two cycles of BEACOPP-based therapy +2 cycles ABVD	PET-4	PET-negative: omitted radiotherapy	RT could be safely omitted for PET-4 negative
SWOG0816	Advanced-stage HL	two cycles ABVD	PET-2	PET-positive: switched to BEACOPP	PET-2 positive patients had good survival but high secondary cancers
HD0801a	Advanced-stage HL	two cycles ABVD	PET-2	PET-positive: IGEV x4 followed by transplant with PET-directed conditioning chemotherapy PET-negative: ABVDx2 with EOT PET negative randomized to radiotherapy	PET-2 positive patients benefited from early treatment intensification with autologous transplant
GITIL 0607	Advanced-stage HL	two cycles ABVD	PET-2	PET-positive: randomized to BEACOPP-based chemotherapy ± Rituximab PET-negative: randomized to ABVDx4 ± RT for nodal disease	PET-2 based escalation to BEACOPP is effective for high risk patients
RATHL	Advanced-stage HL	two cycles of ABVD	PET-2	PET-positive: consolidative RT, highly avid PET switched to BEACOPP PET-negative: randomized to AVD or ABVD	PET-2-negative patients had lower pulmonary toxicity
HD15	Advanced-stage HL	6–8 cycles of BEACOPP-based therapy	EOT PET	PET-positive: radiotherapy	persistent EOT PET-negative mass had a similar prognosis to CR patients
HD18	Advanced-stage HL	two cycles of BEACOPP-based therapy	PET-2	PET-positive: randomized to more therapy ± Rituximab PET-negative: randomized to less therapy	PET2-negative patients can have reduced cycles of chemotherapy
OPTIMAL >60	Aggressive NHL,>60 years old	four cycles of R-CHOP or R-CHLIP-14	PET-4	PET-positive: additional chemoradiotherapy PET-negative: 4xR	Chemoradiotherapy can be spared with PET-4 negative
PETAL	Aggressive NHL	two cycles of CHOP +Rituximab	PET-2	PET-positive: R-CHOPx6 PET-negative: R-CHOPx4	PET-2 predicted survival but treatment intensification did not improve outcomes
GELA LNH073B	Aggressive NHL	two and 4 cycles of R-ACVBP14 or R-CHOP14	PET-2 and PET-4	PET-2 and PET-4 negative: standard PET-2 positive and PET-4 negative: high dose chemotherapy and ASCT PET-4 positive: salvage followed by ASCT	PET-4 positivity had lower EFS, PET-4-negative was similar, change in SUV max associated with improved outcomes

^aNot fully published

The future of imaging: radiomics

In the era of precision medicine, the ability to extract high-dimensional data from clinical imaging in order to understand biology and guide therapy, or radiomics, has emerged as a promising area of research. As is the case with genomics studies, radiomics studies assume that the radiologic and functional imaging properties of tumors correlate with the tumor biology is a measurable way. There are multiple tools to allow extraction of simple and complex radiomic features from clinical imaging by analyzing each voxel within an image; however, the analytic methods for understanding the clinical significance of radiomic features in cancer is still an area of active research. Many of the features that are routinely studied in radiomics research are actually established in the literature, but the recent renaissance of radiomics is attributed to better surrogate endpoint comparisons and modern computing power.¹²³ Radiomic features are summarized as shape features (e.g., volume), first-order features based on individual voxels, second-order features accounting for the intervoxel relationships, and higher order features involving advanced mathematical features (Table 4).

Table 4.

Radiomic Features

Class	Example Features
Size and Shape	volume, maximum diameter, surface, compactness, sphericity
First Order	Voxel mean, median, maximum, minimum, skewness, kurtosis, uniformity, entropy
Second Order	Grey-level co-occurrence matrix, run length matrix, size zone matrix
High Order	fractal analysis, wavelet transform

The state of the field of radiomics is reviewed well elsewhere.¹²⁴ Simple radiomic features like maximum SUV are demonstrated to correlate with aggressiveness¹²⁵ and response to chemotherapy in lymphomas.^126,127 Metabolic tumor volume, or the metabolically active tumor defined from the FDG PET, has been shown to be prognostic when measured at the baseline.¹²⁸ Radiomic features are demonstrated to correlate with response to chemotherapy in Hodgkin’s and non-Hodgkin’s lymphoma, although primarily with aggressive lymphomas.^129–132

Radiomics analysis can be broken down into image acquisition and reconstruction, image segmentation and rendering, feature extraction and feature qualification and databases and data sharing, each of which has unique challenges.¹³³ More recently, researchers have turned a critical eye toward retrospective radiomics studies that do not account for the many shortcomings of the approach.¹³⁴ Currently, efforts are underway to promote standardization, to improve the reliability and generalizability of radiomics research.^135,136

Conclusion

The journey from early drawings of Hodgkin’s patients to deep learning with radiomics in lymphoma has taken nearly 200 years, and in many ways, it parallels the journey of medicine. We are now able to identify involved lymph nodes and body sites both before and after treatment to offer patients improved treatment outcomes. As imaging methods continue to improve sensitivity, we will be able to use personalized medicine approaches to give targeted and highly focused therapies at even earlier time points, and ideally, we can obtain long-term disease control and cures for lymphomas.