OUP user menu

Journal of Crohn's and Colitis: 10 (10)


Laurence J. Egan, Ireland

Associate Editors

Shomron Ben-Horin, IsraelSilvio Danese, ItalyPeter Lakatos, HungaryMiles Parkes, UKJesús Rivera-Nieves, USABritta Siegmund, GermanyGijs van den Brink, NLSéverine Vermeire, Belgium


Published on behalf of

Effects of infliximab therapy on transmural lesions as assessed by magnetic resonance enteroclysis in patients with ileal Crohn's disease


Background and aims: Anti TNF therapy induces mucosal healing in patients with Crohn's disease, but the effects on transmural inflammation in the ileum are not well understood. Magnetic resonance-enteroclysis (MRE) offers excellent imaging of transmural and peri-enteric lesions in Crohn's ileitis and we aimed to study its responsiveness to anti TNF therapy.

Methods: In this multi-center prospective trial, anti TNF naïve patients with ileal Crohn's disease and with increased CRP and contrast enhanced wall thickening received infliximab 5 mg/kg at weeks 0, 2 and 6, and q8 weeks maintenance MRE was performed at baseline, 2 weeks and 6 months and assessed based on a predefined MRE score of severity in ileal Crohn's Disease.

Results: Twenty patients were included; of those, 18 patients underwent MRE at week 2 and 15 patients at weeks 2 and 26 as scheduled. Inflammatory components of the MRE index decreased by ≥ 2 points and by ≥ 50% at week 26 (primary endpoint) in 40% and 32% of patients (per protocol and intention to treat analysis, respectively). The MRE index improved in 44% at week 2 and in 80% at week 26. Complete absence of inflammatory lesions was observed in 0/18 at week 2 and 13% (2/15) at week 26. The obstructive elements did not change. Clinical and CRP improvement occurred as early as wk 2, but only CDAI correlated with the MRE index.

Conclusion: Improvement of MRE occurs from 2 weeks after infliximab therapy onwards and correlates with clinical response but normalization of MRE is rare.

  • Crohn's disease
  • Imaging
  • Complications
  • Anti-TNF therapy
  • MR enteroclysis

1 Introduction

The proinflammatory cytokine tumor necrosis factor-alpha (TNFα) plays an important role in the pathogenesis of Crohn's disease, a chronic inflammatory disorder of the gastrointestinal tract. Infliximab, an anti-TNFα monoclonal antibody, binds to TNFα with high affinity, thereby neutralizing its biological activity. The administration of infliximab to patients with moderately-to-severely active CD disease induces and maintains clinical remission and promotes mucosal healing. Mucosal healing with infliximab is associated with a decrease in disease related hospitalizations.1 However CD is a transmural disease; intestinal wall thickening with fibrosis, muscular hypertrophy and mesenteric hypertrophy with fat accumulation and hypervascularization are characteristic features of CD. The influence of infliximab on transmural CD is unclear. Although initial case reports of intestinal obstruction early after infliximab therapy for ileal CD generated concern about the safety of infliximab in ileal CD with an inflammatory stenosis, a recent analysis of prospective registry data suggests that the risk of obstruction is not specifically increased in infliximab treated patients.2,3 Transmural imaging with Computed tomography (CT) or magnetic resonance (MR) enterography or enteroclysis provides appropriate assessment of transmural lesions and extramural complications.4 Magnetic Resonance enteroclysis (MRE) offers unique imaging of the transmural inflammation in intestinal CD and contrast uptake behavior of the diseased bowel wall may allow the discrimination of inflammatory from fibrotic lesions.516 The discriminative power of imaging is however, limited since Cd is probably never purely fibrostenotic and features such as contrast enhancement are also influenced by fibrotic changes.15 Bowel filling via a jejunal tube within the magnet under MR fluoroscopic monitoring also allows for controlled and reliable bowel distension. Recent data with MRE indicate that this technique is very sensitive to detect both inflammatory and fibrotic changes.11

Therefore, An Open Label, Prospective, Multi-Center Trial, on the Effect of the Anti-TNFα Chimeric Monoclonal Antibody Infliximab on Inflammatory and Fibrous Lesions in Patients with Intestinal CD (ACTIF) was designed to assess the impact of infliximab on transmural lesions of CD as visualized by MRE.

2 Patients and methods

This was an open label, prospective, multi-center pilot trial in patients with documented ileal or ileocolonic Crohn's disease with an indication to start infliximab (Remicade®, Janssen Biotech Inc., Horsham, PA, USA) 5 mg/kg IV induction at weeks 0, 2 and 6 and maintenance every 8 weeks thereafter. The indication to start anti TNF therapy for luminal Crohn's disease was based on the European label and patients had moderate to severe active CD failing steroid and/or immunosuppressive therapies. To qualify for the study patients had to be ≥ 18 and ≤ 65 years of age, with a CDAI > 220 and a high sensitivity CRP (hs-CRP) of more than 5 mg/l. Imaging criteria for inclusion were markedly increased Gadolinium (Gd) uptake of the small bowel wall and ileal wall thickening on baseline contrast enhanced MRE. Exclusion criteria were active or latent tuberculosis, contraindications for MRE, treatment with more than 15 mg of systemic corticosteroids (prednisone equivalent) within the 2 weeks prior to baseline MRE, prior bowel resection of > 100 cm or documented abdominal abscess or internal fistula as well as medical contraindications for anti TNF therapy. A standard corticosteroid taper was enforced and all other Crohn's disease therapies were kept stable during the first 26 weeks. Initiation of any medical therapy to treat Crohn's disease was not allowed. After obtaining written informed consent, MRE was scheduled and performed prior to the first infliximab infusion, at weeks 2 and 26. If a MRE had been performed within one month from starting treatment (adhering to the predefined and detailed imaging protocol instructions and without subsequent changes in therapy), the results of this MRE could be used for baseline scoring.

The Crohn's disease activity index, CDAI,17 Harvey-Bradshaw index, HBI,18 the IBD (Inflammatory Bowel Disease) Questionnaire,19 concomitant medications and adverse events were recorded at every visit during the 26-weeks duration of the trial. After the main study of 26 weeks, patients were followed for safety and efficacy of infliximab until 2 years from inclusion.

Contrast enhanced MRE with intraluminal and intravenous contrast application was performed on 1.5 Tesla (T) MR scanners using an identical imaging protocol and imaging parameters at all institutions (Suppl. Table 1). A detailed MRE study manual was developed and distributed by the Institute of Clinical Radiology, University of Munich, (KAH) including comprehensive instructions for appropriate patient preparation, imaging technique, sequences and imaging parameters. In addition, at least one dedicated abdominal imaging specialist from each participating center received a one-day hands-on training at the training site at University of Munich, prior to initiation of the site. An MRE performed per proctocol in a patient with suspected ileal CD was sent for central review to the training center in Munich (KAH) in order to ascertain appropriate technique and imaging quality and to approve final initiation of the site (Fig. 1).

Figure 1

MICD scores correlated with CDAI but not with CRP. The graph represents the data for the total MICD score only. MICD scores, and CDAI and CRP values were pooled at all time points.

For MRE all patients underwent placement of a naso-jejunal tube prior to the MRI examination. Final tube position with the tip beyond the ligament of Treitz was confirmed with fluoroscopy. Patients were transferred to the MRI unit and positioned in supine position in the MRI bore. The naso-jejunal tube was connected to an automated infusion pump via an infusion system and 1000–2500 ml of 0.5% methylcellulose solution were applied at a speed of 80–100 ml/min until adequate and homogeneous distention of all small bowel segments was obtained. The filling process was monitored using repetitive SSFSE imaging every 1–2 min during infusion (HASTE-online; see Table 1). Standard protocol imaging was started after completion of the filling phase with SSFSE and SSFP in two planes followed by non enhanced T1w 3D GRE (See supplementary Table 1). Prior to non enhanced T1w 3D GRE, spasmolytic agents (Butylscopolamine, Buscopan ®, 20–40 mg, Boehringer Ingelheim, Germany) were administered to control bowel motion. For subsequent contrast enhanced imaging an intravenous contrast agent at standard dose (0.1 mmol/kg/BW; Magnevist ®, Bayer Schering, Germany) was injected at 1.5 ml/s and T1w 3D GRE was performed approximately 60 s, 90 s and 180 s post injection. Protocol was completed with T1w 2D GRE imaging (Supplementary Table 1).

View this table:
Table 1

Baseline characteristics of the patients.

N = 20
Age (median, range)29 (19–54)
Disease duration (median, years)4.8 (0.3–21.0)
CDAI277 (245–317)
CRP (mg/l)15.3 (6.0–25.0)
HBI9.0 (7.3–10.0)
Weight (kg)65 (57–80)
MICD7 (5–9)
Concomitant medicationsAza/6-MP: 11/20 Systemic steroids: 4/20 Budesonide: 7/20 5-ASA: 3/20
  • Note: All patients in the safety population (n = 20) were included. Data are median and numbers in parentheses indicate range.

A newly developed index, MRE score of severity in Ileal Crohn's Disease (MICD) was used to score CD severity and complications. The MICD was designed to be a combined index of transmural inflammation, extramural involvement and signs of obstructive disease (Suppl. Fig. 2).

The MICD index ranges from 0 to 14 with the inflammatory scores ranging from 0 to 8 and the obstructive scores from 0 to 6. The criteria for scoring lesions associated with transmural ileal CD lesions were based on an extensive review of the recent literature and on a Consensus between Gastroenterologists and Radiologists of this study group.

Scoring included the assessment of the presence and degree of active inflammation as well as the presence and degree of obstruction. Signs of inflammation were determined as increased thickness of the bowel wall, augmented contrast enhancement after intravenous administration of Gd-DTPA and inflammatory reactions beyond the intestinal wall in the adjacent mesenteric tissue as a sign of extramural disease. Measurements of the small bowel wall thickness were performed on T2w weighed and SSFP images at three different levels of the inflamed bowel segment where wall thickening was subjectively deemed as maximal. As in contrast enhanced studies the depiction of the wall dimensions depends heavily on the uptake pattern, these are less reliable for the exact delineation of the bowel wall. Distance measurements were placed in cross sectional images where the small bowel was perpendicular to the imaging plane. Of the three measurements in each sequence the median was calculated as a basis for the scoring. Zoom function was allowed to facilitate a more precise placement of measurements. In order to assess the contrast enhancement of the bowel wall, first, “normal” bowel wall enhancement was determined for reference. Hereto, regions of interest (ROI) for signal intensity (SI) were placed in one location of normal bowel and subsequently in three different locations of diseased bowel and in both the unenhanced and enhanced T1w images to calculate the relative uptake in percentages. The obstructive component was assessed by measuring the minimal diameter of the bowel lumen over all sequences in a cross sectional plane strictly perpendicular to long axis of the bowel and the degree of prestenotic dilatation upstream to the diseased bowel segment. Prestenotic dilatation was semi-quantified taking the maximum diameter of dilatation and comparing it to the luminal distension of unaffected bowel of the same anatomic region.

At the end of the trial all MRE images were de-identified as for patient ID, patient number, date of assessment or visit number and distributed among several radiologists for review and scoring (KAH, DT, DV, PM). After all MRE had been read, the scores were unblinded and correlated to clinical outcomes.

The study protocol was approved by the Institutional Review Boards of all participating centers.

3 Endpoints and statistical considerations

The primary endpoint of this study was the number of patients achieving a clinically significant change in the MICD score defined as an overall improvement of at least 2 points and an improvement of at least 50% in the inflammatory subscores (range 0–8) at wk 26. For the primary endpoint we focused on a minimal 50% improvement of the inflammatory subscores as we assumed they would best respond to an anti-inflammatory treatment. Secondary endpoints included change in MICD total and individual scores over time, the number of patients with clinical improvement and CDAI remission (< 150), the correlation between MICD scores and CRP and CDAI respectively and the safety of infliximab in this population particularly pertaining to intestinal stenosis and resections. Non-parametric tests were used to analyze changes in continuous variables and Spearman's rank statistics were used to test for correlations.

Both an intention-to-treat (ITT) and per protocol analysis were planned in the assessment of primary and secondary efficacy endpoints and for safety outcomes. The ITT population included all patients who received one or more IFX infusions and did not violate the inclusion criteria. The per-protocol population included all ITT patients where an MICD score was available at weeks 0, 2 and 26 and IFX dosing had not been interrupted prior to wk 26.

Last Observation Carried Forward (LOCF) with non-responder imputation, was used in the ITT analysis. This means that patients with missing MRE data were considered to be non-responders for the primary endpoint at wk 26. We performed no formal power calculations, since this was a pilot trial and no literature data on the response of transmural lesions to anti TNF therapy were available when this study was designed.

4 Results

4.1 Patient enrollment

From July 2007 until September 2010, 20 patients (13 women and 7 men, median age 29 (19–54), median disease duration 4.8 years (0.3–21.0)) were enrolled in the study at the different sites and formed the study population. Patient characteristics are summarized in Table 1. One patient was excluded because of a major protocol violation (normal CRP at entry, ITT: n = 19) and 15 patients fulfilled all criteria for the per protocol analysis (Suppl. Fig. 2).

4.2 MRE response

A decrease in MICD index with at least 2 points or more and an improvement with at least 50% in the inflammatory subscores on the wk 26 MICD score (primary endpoint) was reached by 40% (6/15) of patients per protocol and 32% (6/19, ITT analysis) respectively. The median total MICD scores improved at week 26 compared to baseline [wk 26: 5 (IQR: 3–6.8), wk 0: 7 (5–9), p < 0.03] as did the inflammatory sub-scores [wk 26: 3 (IQR: 0.25–4), wk 0: 4 (3–6), p < 0.03] (Figs. 2 and 3). In contrast, the obstructive disease sub-scores were similar at the different time points (Table 2). Absence of inflammatory lesions (as defined by a score of 0 in the inflammatory sub-scores of the MICD index) occurred in none of the patients at wk 0 and in 2/15 patients at wk 26. Worsening of the scores with a maximum of 1 point was observed in one patient for the inflammatory scores and in one patient for the obstructive disease scores (data not shown).

Figure 2

Representative images of the evolution of the MRE before and 2 weeks after infliximab 5 mg/kg IV. Coronal and transverse sections of the same patient are depicted. Arrows focus on the ileal segment that was selected for assessment.

Figure 3

Change of the inflammatory components of the MICD and of the total score in individual patients.

View this table:
Table 2

Clinical and MRE outcomes after treatment.

Wk 0 (n = 19) median, IQRWk 2Wk 26
Total MICD score7.0 (5.0–9.0)6.5 (6.5–8.8)5.0 (3.0–6.8)
Inflammation subscores4.0 (3.0–6.0)4.0 (3.0–5.5)3.0 (0.25–4.0)
Obstructive subscores3.0 (3.0–4.0)3.0 (3.0–4.0)3.0 (2.0–3.0)
CDAI277 (249–338)169 (109–269)126 (67–180)
CRP (mg/l)15.3 (6.0–25.0)3.0 (1.8–5.6)6.6 (5.5–12.3)
HBI9.0 (7.3–10.0)5.0 (3.3–9.3)4 (2.0–6.0)°
IBDQ138 (113–162)157 (131–168)171 (139–191)
Weight (kg)65 (57–80)65 (59–82)69 (59–81)

Note: all patients in the ITT population (n = 19) were included.

  • p < 0.05.

  • ° p < 0.01.

4.3 Clinical outcomes and correlation with MRE response

The median CDAI, median HBI and hs-CRP improved as early as week 2 (Table 2). At week 2, 8/19 patients had achieved CDAI remission (< 150) and 6/19 HBI remission (< 4); at wk 26 remission rates were 6/19 for CDAI and 7/19 for HBI. C-reactive protein levels dropped by at least 50% compared to baseline in 14/18 patients at wk 2 and in 8/15 at wk 26. The individual MICD total scores and inflammatory sub-scores correlated with the CDAI values (p < 0.001 and p < 0.01), but not with CRP. (Table 3; Fig. 1). A numerical but non-significant increase in IBDQ was observed over time (Table 2). The median weight was unaffected by treatment and the median change in weight was + 1.4 (0–4) kg.

View this table:
Table 3

Correlation of MICD with CRP and CDAI.

Total MICDR = 0.52 (p = 0.0003)R = 0.12 (p = 0.4)
Inflammation subscoresR = 0.45 (p < 0.01)R = 0.13 (p = 0.4)
  • Note: Spearman Rank statistics were used to test for potential correlation between all paired data at different timepoints.

4.4 Safety

Through week 26 five serious adverse events were reported in 5 patients and these were judged to be possibly related to infliximab therapy. Three patients had discontinued infliximab by wk 26 due to pregnancy (n = 1), an infusion reaction (n = 1) and pancreatitis (n = 1, concomitant azathioprine) (Suppl. Table 2).

No CD related surgeries were needed until wk 26, although one patient experienced a pyloric stenosis that was endoscopically dilated. Beyond wk 26 until the end of the two year follow up two more ileo-colonic symptomatic stenoses were observed but none of the patients required surgery. Seven more patients stopped infliximab beyond wk 26, mainly due to side effects (infusion related reactions n = 4, skin lesions n = 1, loss of response n = 2). There were no serious adverse events related to the MRE.

5 Discussion

The effect of infliximab and other anti TNF agents on mucosal lesions has been well studied, but the impact of anti TNF agents and medical treatment in general on transmural and mesenteric lesions is less clear. We here provide the first prospective evidence that the anti TNF agent infliximab induces rapid improvement of transmural inflammation assessed by MRE in an open label prospective study. Interestingly, complete resolution of lesions was a rare phenomenon and the vast majority of patients had persistent signs of inflammation and signs of obstruction at imaging did not change with infliximab therapy for 26 weeks. Since inflammation leads to fibrostenosis in Crohn's disease,20,21 one can speculate that earlier introduction of thiopurines and anti TNF therapies may prevent damage of the affected bowel segments.

Recently, results of a retrospective analysis of a cohort of patients treated with infliximab for Crohn's disease and followed with CT enterography at the Mayo Clinic indicated that signs of active Crohn's disease such as wall thickening and contrast enhancement, but more importantly signs of mesenteric inflammation (comb sign) improved in more than 60% of patients with treatment. The authors suggested that they had underestimated the beneficial effects of the anti TNF therapy since repeated CT had been performed only in patients with ongoing symptoms. We however, confirm with a prospective trial that even in patients with a clear clinical response to infliximab therapy, persistence of transmural abnormality is common using MRE. Pre-existing bowel damage due to longstanding inflammation and subsequent fibrostenosis most likely contributes to the fact that in our study the sub-scores aimed at assessing obstructive disease were least improved by anti TNF therapy. However, some signs of inflammation such as contrast enhancement and extramural involvement persisted in the majority of patients and a complete resolution of lesions was not observed. It is possible that more delayed MRE imaging beyond wk 26 after the start of infliximab therapy would have demonstrated additional improvement in MICD scores, but this was not assessed. Therefore, data in a larger set of patients with a long term follow up are needed to study the clinical relevance of evaluating signs of obstructive disease on MRE. The impact of persistent lesions on long term outcomes also deserves future exploration. However, it is important to note that, albeit we only followed 19 patients out to 2 years after start of infliximab, no surgeries for obstructive disease were needed.

We opted for enteroclysis in this study since this technique is considered to provide better distension of small bowel loops and assessment of stenotic disease than standard enterography. However, recent evidence suggests the adequate filling can be obtained with MR enterography.11,12 We also chose to design a novel MRE based score to quantify the results of our assessments. Based on the available literature at the time of study initiation and clinical experience, items such as wall thickening, increased contrast enhancement, extramural involvement and pre-stenotic dilation were selected and prioritized. Although one strong indicator (submucosal edema with increased signal intensity in T2 weighted imaging) was not included in our score, all other signs have been validated in the literature in the meantime and were confirmed as strongly indicative of active inflammatory disease.1012,14

Although we opted for a primary endpoint at six months, a second MRE was performed as early as 2 weeks after the initiation of treatment, since infliximab induces early clinical improvement in patients with luminal Crohn's disease.22 Although some degree of improvement in transmural inflammation was observed at week 2, clinical (CDAI and HBI) and CRP improvement was more pronounced at this early time point. We hypothesize that a rapid improvement of mucosal rather than transmural inflammation drives the early change in symptoms and in CRP. However, earlier data with CT enterography have indicated that persistent serosal and extraintestinal inflammation was associated with an increased CRP.23 In this study, we observed a poor correlation with CRP in general.

Our study has several limitations. This was a pilot trial in a limited number of patients. Since this study is the first prospective trial on the early and medium term responsiveness of MRE to medical therapy in ileal Crohn's disease and since we opted for an open label strategy with infliximab, formal power calculation were not feasible. Our results should therefore be interepreted with caution. Patients were not consented to a baseline ileoscopy and therefore we are unable to correlate the MRE finding with mucosal lesions. We used a novel MRE based activity score for ileal Crohn's disease that has not been externally validated. However, the results of our study indicate that the MICD index is sensitive to change over time. We also found a moderate correlation with clinical activity assessed by CDAI but not with CRP. In the Mayo cohort study concordance between CT enterography response and clinical improvement was poor, but clinical outcomes were not fully quantified.13 Since 2006 more data on the use of MRI to assess ileocolonic Crohn's disease have become available1012 and a ‘segment based’ quantitative score applicable to the terminal ileum and the colon has been validated.10,11 Most of this evidence was obtained with MR enterography, but the criteria proposed in the cohort studies and in the validated MaRIA score10,11 correspond well with our MICD score. We were unable to change the primary outcome of our study after it had been initiated in 2007, but further cross validation of the different MRI based activity scores is needed.

Although it was an open label study, MRE evaluations were carefully and adequately blinded and we observed a correlation between clinical activity and the MICD. In addition, whilst there were a limited number of participants, this was intended as a pilot observational study. Finally, the broad range of disease duration (Median 4.8 years, range 0.3–21 years) in this study group may have biased towards a poor response.

We selected patients with active inflammation based on the presence of an increased CRP and a marked increase in contrast uptake by the ileal wall. Patients with evidence of abdominal sepsis (abdominal abscess) were excluded since anti TNF therapy would be contraindicated. The selection of patients based on signs of active transmural inflammation stemmed from the theoretical concern that anti TNF therapy could precipitate surgery in patients with an ileal stenosis. However recent data have confirmed that patients with an increased baseline CRP respond better to infliximab therapy24,25 and in the selected population entered in our trial, no intestinal surgery was needed until 6 months after the start of infliximab therapy despite the presence of ileal wall thickening and some luminal dilation in all patients (though one patient developed a symptomatic pyloric stenosis that was successfully dilated). Finally, our sample size was too small to allow for a proper identification of predictors of response based on cross sectional imaging and further larger studies are awaited.

In summary, infliximab induces improvement of transmural and extraintestinal inflammatory lesions in ileal Crohn's disease assessed by MRE but signs of inflammation and obstruction persist despite clinical remission. More prospective, controlled studies are needed to better define the role of transmural imaging to evaluate the efficacy of new therapeutic agents and to delineate the clinical consequences of transmural healing or persistence of inflammatory lesions.

Conflicts of interest

Gert Van Assche, Thomas Ochsenkühn, Edouard Louis, Jean-Frédéric Colombel, Simon Everett, Paul Rutgeerts and Séverine Vermeire served as consultants for or received speakers fees from Schering–Plough, a subsidiary of Merck & Co, Inc. or Centocor Research & Development, a division of Johnson & Johnson Pharmaceutical Research & Development, LLC.

Jean-François Rahier received Research Support from Schering–Plough, a subsidiary of Merck & Co.

Dirk Vanbeckevoort, Isolde Aerden, Olivier Ernst, Karin Herrmann, Paul Meunier and Damian Tolan have no conflicts of interest to declare.

Alessandra Oortwijn is an employee of Centocor Research & Development, a division of Johnson & Johnson Pharmaceutical Research & Development, LLC.

Appendix A Supplementary data

Supplementary material.

Appendix A Supplementary data

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.crohns.2013.01.011.


  • Gert Van Assche, Karin Herrmann, Alessandra Oortwijn and Thomas Ochsenkühn designed the study. Karin Herrmann developed the MR enteroclysis protocol and performed the site training. Gert Van Assche, Edouard Louis, Simon Everett, Jean-Frédéric Colombel, Jean-François Rahier, Séverine Vermeire, Paul Rutgeerts and Thomas Ochsenkühn recruited patients and contributed to the writing of the manuscript. Dirk Vanbeckevoort, Karin Hermman, Paul Meunier, Damian Tolan, Olivier Ernst reviewed the MR images and contributed to the writing of the manuscript.

  • ☆☆ The corresponding author had full access to all of the data and takes full responsibility for the veracity of the data and statistical analysis.

  • Role of funding source: This trial was supported by a restricted grant from Centocor BV Leiden. The authors independently designed the study, analyzed the data and wrote the manuscript.

  • 1 Gert Van Assche and Karin Herrmann contributed equally to this paper.


View Abstract