The study, published in BMC Medical Imaging, addresses a long-standing tension in lymphoma staging. Diffuse large B-cell lymphoma, the most common aggressive non-Hodgkin lymphoma in adults, is staged in part by determining whether the disease has spread to the bone marrow. For decades, the only way to answer that question definitively was the iliac crest biopsy, a procedure that samples just a tiny fraction of the total marrow volume. Meanwhile, positron emission tomography using the radioactive glucose analog fluorodeoxyglucose offers a whole-body view of metabolic activity, because rapidly dividing lymphoma cells consume glucose voraciously and light up on the scan. The problem has always been knowing when a bright signal in the marrow truly means cancer rather than something benign.

The Korean team, led by Sang Jun Byun and Yunhwi Hwang, who contributed equally to the work, with corresponding author Byungwook Choi, proposed an elegantly simple solution: use the liver as an internal brightness yardstick. Two nuclear medicine physicians independently classified the uptake of the radioactive tracer in both the bone marrow and the spleen as either High or Low relative to the liver’s own signal. If either organ appeared brighter than the liver — or if either showed focal, localized hot spots — the case was labeled “Any High.” If both organs stayed dimmer than the liver, the case was “Both Low.” The logic is rooted in lymphoma biology: the spleen, like the marrow, is a lymphoid organ where disseminated disease tends to lodge, so simultaneous metabolic activation in both sites is far more suspicious than activation in either alone.

To test the rule, the researchers assembled a primary diagnostic cohort of 107 patients with newly diagnosed diffuse large B-cell lymphoma who had undergone both pretreatment PET/CT and adequate bone marrow biopsy. The biopsy came back positive in 18 of those patients. When the Any High classification was applied, it correctly identified 83.3 percent of the true positive cases — a sensitivity with a 95 percent confidence interval running from 58.6 to 96.4 percent. More importantly for patients hoping to avoid the needle, the rule correctly cleared 80.9 percent of the true negatives, yielding a specificity of 71.2 to 88.5 percent. The negative predictive value — the probability that a Both Low result truly means the marrow is clean — reached an extraordinary 96.0 percent, with a confidence interval of 88.8 to 99.2 percent. Overall accuracy was 81.3 percent.

That negative predictive value is the number that matters most clinically. In this cohort, only 3 of the 75 patients classified as Both Low actually harbored marrow involvement. In other words, if a patient’s PET scan showed subdued marrow and splenic activity relative to the liver, there was a 96 percent chance the biopsy would find nothing. For the remaining small fraction, the biopsy would still catch the disease, but for the overwhelming majority, the painful procedure could potentially be skipped — pending the prospective validation the authors themselves insist is required before any change to clinical practice.

The researchers did not stop at the combined rule. They compared it against the simpler alternative of looking at the bone marrow alone. The combined Any High approach traded some specificity for sensitivity: it caught more true positives but flagged more false alarms. Yet the overall accuracy was statistically indistinguishable — 81.3 percent for the combined rule versus 82.2 percent for marrow alone, with an exact McNemar p value greater than 0.999. The team also ran a sensitivity analysis excluding patients whose marrow showed focal uptake, reducing the cohort to 94 patients with 12 biopsy-positive cases, and confirmed the estimates held. They even stress-tested the results by reclassifying or excluding two indeterminate biopsies, and the numbers remained robust.

Reproducibility is where many visual classification schemes in nuclear medicine collapse, but this one held up well. Inter-reader agreement between the two independent physicians was substantial to almost perfect for the principal criteria, with Cohen’s kappa values ranging from 0.750 to 0.843. The one soft spot was background splenic uptake, which showed lower agreement at kappa 0.648, compared with 0.773 for background marrow uptake. The spleen, it turns out, is a trickier organ to grade because its normal tracer uptake sits closer to the liver’s baseline, making the visual distinction more subtle. Still, the authors note that the combined classification itself — the decision that ultimately drives the clinical call — was highly reproducible.

Beyond diagnosis, the team explored whether the PET-based classification carried prognostic weight. In a separate cohort of 141 patients, they tracked progression-free survival and overall survival using Cox regression models. Over the follow-up period, 66 progression events and 49 deaths occurred. When adjusted for the International Prognostic Index — the standard clinical risk score incorporating factors such as age, stage, lactate dehydrogenase, and performance status — patients classified as Any High had roughly twice the hazard of progression or death compared with Both Low patients, with a hazard ratio of 2.04 and a confidence interval of 1.08 to 3.84. Adjustment for the more granular National Comprehensive Cancer Network version of the index yielded a nearly identical result: hazard ratio 1.96, confidence interval 1.05 to 3.69. However, the classification was not independently associated with overall survival, and the authors are careful to frame these survival findings as exploratory rather than practice-changing.

The implications, if validated prospectively, could reshape the staging pathway for thousands of patients each year. Diffuse large B-cell lymphoma is typically treated with the R-CHOP regimen — rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisolone — and accurate staging determines both the intensity of treatment and the radiation fields in some approaches. A reliable noninvasive alternative to marrow biopsy would spare patients a painful procedure, reduce healthcare costs, and accelerate the time from diagnosis to treatment. The liver-referenced approach requires no additional scanning, no new tracers, and no specialized software — just a disciplined visual comparison that any trained nuclear medicine physician can perform on images already acquired as standard of care.

The authors are appropriately measured in their conclusions. They acknowledge that BMB positivity did occur in some Both Low patients, that the study was retrospective and single-centre, and that the survival associations were secondary and exploratory. They explicitly call for prospective external validation before the classification is used clinically. But the core finding stands on its own weight: by reading two organs against one internal reference, a century-old diagnostic dilemma gets a modern, image-based answer that is 96 percent reliable when it says the marrow is clear. For a disease that afflicts tens of thousands of people annually, that is not a small refinement — it is a potential reordering of the very first steps of cancer care.

Subject of Research: Diagnostic performance of integrated bone marrow and splenic FDG PET/CT uptake for predicting bone marrow biopsy positivity in diffuse large B-cell lymphoma

Article Title: Integrated bone marrow and splenic [18F]FDG PET/CT uptake for predicting bone marrow biopsy positivity in diffuse large B-cell lymphoma — a retrospective single-centre study

Article References: Byun, S. J., Hwang, Y., Kim, H. W., & Choi, B. (2026). Integrated bone marrow and splenic [18F]FDG PET/CT uptake for predicting bone marrow biopsy positivity in diffuse large B-cell lymphoma — a retrospective single-centre study. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02761-x

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02761-x

Keywords: diffuse large B-cell lymphoma, FDG PET/CT, bone marrow biopsy, splenic uptake, liver-referenced classification, negative predictive value, inter-reader agreement, lymphoma staging, nuclear medicine, progression-free survival, International Prognostic Index, diagnostic imaging

The two sequencing approaches represent different philosophies for reading genetic material from a patient sample. Targeted next-generation sequencing, often abbreviated tNGS, uses panels of primers or probes to enrich and amplify DNA from a predefined list of pathogens, which concentrates sequencing capacity on organisms that clinicians already suspect. Metagenomic next-generation sequencing, or mNGS, takes the opposite approach: it sequences all nucleic acid in a sample without any preselection, allowing the laboratory to detect virtually any microbe present, including unexpected or fastidious organisms that no one thought to test for. That breadth comes at a cost, both financially and computationally, because vast amounts of background human and environmental DNA must be filtered out before a pathogen signal emerges. The question facing hospitals is whether mNGS’s unbounded scope translates into measurably better care for patients with suspected pulmonary tuberculosis.

To answer that question fairly, the researchers confronted a familiar obstacle in observational medicine: patients are not assigned to tests at random. In routine practice, physicians choose mNGS or tNGS based on how sick a patient appears, what previous tests have shown, and what the patient can afford, which means simple comparisons between the two groups can be badly skewed. The team therefore turned to propensity score matching, a statistical technique that pairs patients from each group who share similar baseline characteristics, effectively simulating the balance of a randomized trial. After this matching procedure, the analysis rested on 403 patients with confirmed pulmonary infections, of whom 304 had undergone tNGS and 99 had received mNGS.

The headline diagnostic finding was one of statistical equivalence. In the matched cohorts, the final rate of tuberculosis diagnosis was 17.4 percent in the tNGS group versus 23.2 percent in the mNGS group, a difference that did not reach statistical significance, with a p value of 0.257. Similarly, the proportion of samples in which the Mycobacterium tuberculosis complex was directly detected by sequencing was 20.4 percent for tNGS and 28.3 percent for mNGS, again without a significant difference at p equals 0.134. When the researchers plotted receiver operating characteristic curves to summarize overall diagnostic accuracy, the area under the curve came to 0.936 for tNGS, with a 95 percent confidence interval of 0.898 to 0.975, and 0.967 for mNGS, with a confidence interval of 0.939 to 0.995. The comparison between these two areas yielded a p value of 1.000, about as close to a tie as statistics allows.

Beneath that tie, however, lay one of the study’s most striking results. Within the matched mNGS cohort of 99 patients, metagenomic sequencing achieved 100 percent sensitivity and a 100 percent negative predictive value for tuberculosis, meaning that in this dataset a negative mNGS result reliably ruled the disease out. That figure deserves careful interpretation: in the larger unmatched full cohort of 104 patients who received mNGS, sensitivity slipped to 95.8 percent and the negative predictive value to 98.7 percent, a reminder that perfect performance in a small matched subset can be fragile. Even so, the authors argue that such a high negative predictive value, if confirmed in larger studies, could make mNGS a powerful tool for excluding tuberculosis in diagnostically challenging cases, where every additional day of uncertainty carries clinical and emotional weight.

The sequencing tests were also benchmarked against the established tuberculosis diagnostic arsenal. In subset-specific area-under-the-curve analyses performed after propensity score matching, sequencing targeting the Mycobacterium tuberculosis complex, referred to as MTBC-NGS, outperformed the interferon-gamma release assay, the tuberculosis DNA test, and acid-fast staining, three workhorses of mycobacterial diagnosis that respectively measure immune response, detect bacterial DNA by conventional amplification, and visualize acid-fast bacilli under the microscope. Notably, MTBC-NGS also showed an advantage over the Xpert MTB/RIF assay, the World Health Organization-endorsed rapid molecular test that has transformed tuberculosis detection worldwide, although that advantage was smaller and only became statistically significant after the researchers applied a Bonferroni correction to account for multiple comparisons. The pattern suggests that while newer sequencing platforms do not render existing tools obsolete, they can extract additional diagnostic signal in the right clinical subsets.

Diagnostic performance, however, is only half the story, because healthcare systems must also weigh what each test costs and how it shapes the patient’s journey through the hospital. Here the study delivered a clear verdict: the mNGS group incurred substantially higher healthcare costs than the tNGS group. Metagenomic sequencing’s untargeted design requires deeper sequencing, more intensive bioinformatics, and greater laboratory infrastructure, expenses that are inevitably passed on to patients and payers. Targeted sequencing, by focusing resources on a curated panel of pathogens, achieves comparable diagnostic accuracy for tuberculosis at a lower price point, a distinction that matters enormously in the high-burden settings where tuberculosis testing decisions are made most often.

One widely held assumption did not survive the study’s statistical scrutiny. Earlier comparisons had suggested that patients who received mNGS experienced longer hospital stays, an observation that seemed intuitive given the turnaround time and complexity of untargeted sequencing. But when the researchers adjusted their analysis for the calendar period in which patients were tested, that difference in length of stay lost its statistical significance. The authors conclude that the apparent prolongation of hospitalization may have been attributable in part to temporal changes in clinical practice over the course of the study rather than to the test modality itself. It is a subtle but important methodological lesson: in fast-evolving fields like clinical genomics, when a test was ordered can matter as much as which test was ordered.

The study, led by Xinle Liang and colleagues in the Department of Respiratory and Critical Care Medicine at Yan’an University Affiliated Hospital, was approved by the hospital’s ethics committee and conducted in accordance with the Declaration of Helsinki. The team acknowledged the Xi’an Regional Medical Laboratory Center, which provided both tNGS and mNGS testing services, and the hospital’s clinical laboratory for supporting the work. As a retrospective, single-center analysis, the investigation carries inherent limitations that the authors themselves flag, most notably the modest size of the mNGS group even after matching, and the possibility that unmeasured differences between patients influenced both test selection and outcomes. The 100 percent sensitivity and negative predictive value observed in the matched mNGS cohort, in particular, is explicitly presented as a finding that requires validation in larger studies before it can guide practice.

For clinicians wrestling with suspected pulmonary tuberculosis, the practical message is one of complementarity rather than replacement. Metagenomic sequencing offers unmatched breadth and, in this cohort, an exceptional ability to rule tuberculosis out when results are negative, making it attractive for critically ill or diagnostically perplexing patients in whom the causative pathogen could be almost anything. Targeted sequencing delivers essentially equivalent accuracy for tuberculosis at lower cost, positioning it as the rational first-line genomic option when the clinical question is narrower. As sequencing costs continue to fall and bioinformatics pipelines mature, studies like this one, grounded in real-world propensity-matched cohorts rather than idealized trial conditions, will be essential for deciding which patients truly benefit from the most expensive diagnostic technology and which can be served just as well by its leaner cousin.

Subject of Research: Comparative diagnostic performance of targeted and metagenomic next-generation sequencing for pulmonary tuberculosis

Article Title: Clinical value of targeted next-generation sequencing and metagenomic next-generation sequencing in the diagnosis of pulmonary tuberculosis: a real-world propensity score matching study

Article References: Liang, X., Sun, Y., Sun, J., Zhao, J., He, F., & Bai, Z. (2026). Clinical value of targeted next-generation sequencing and metagenomic next-generation sequencing in the diagnosis of pulmonary tuberculosis: a real-world propensity score matching study. BMC Infectious Diseases. https://doi.org/10.1186/s12879-026-14537-3

Image Credits: AI Generated

DOI: 10.1186/s12879-026-14537-3

Keywords: pulmonary tuberculosis, targeted next-generation sequencing, metagenomic next-generation sequencing, propensity score matching, Mycobacterium tuberculosis complex, Xpert MTB/RIF, diagnostic accuracy, negative predictive value, healthcare costs, BMC Infectious Diseases, molecular diagnostics, retrospective study

Hepatocellular carcinoma, the most common form of primary liver cancer, is a disease of vigilance. Even when a tumor is successfully eliminated, the underlying cirrhotic liver that produced it remains fertile ground for new malignancies, and recurrence can appear either at the ablation site or anywhere else in the organ. International guidelines from the American Association for the Study of Liver Diseases, the European Association for the Study of the Liver, and the Korean Liver Cancer Association therefore recommend serial imaging every three to six months after treatment, a schedule patients typically maintain indefinitely. Missing an early recurrence can mean the difference between another curative intervention and advanced, incurable disease.

The standard surveillance toolkit has long been a compromise. Computed tomography with intravenous contrast exposes patients to cumulative radiation and carries risks of contrast allergy and kidney injury, making it a suboptimal choice for patients who may need dozens of scans over a lifetime. Complete-protocol MRI with gadolinium-based contrast is widely considered the best imaging modality for detecting and characterizing hepatocellular carcinoma, but it is constrained by long scan times, high cost, limited scanner access, and the specialized expertise required to interpret dozens of image series. That bottleneck has motivated radiologists to ask a deceptively simple question: how much of the full scan is actually necessary?

The UCLA team, led by Pedram Keshavarz and senior authors Steven S. Raman and David S. K. Lu, tested an abbreviated MRI protocol built around just two sequences: the hepatobiliary phase of a gadoxetic acid-enhanced scan and a single-shot T2-weighted sequence. Gadoxetic acid is a liver-specific contrast agent, roughly half of which is excreted through the bile after injection. In the late hepatobiliary phase, typically captured fifteen to twenty minutes after administration, functioning hepatocytes take up the agent through organic anion transporting polypeptide transporters on their surface and appear bright. Most hepatocellular carcinomas lack these transporters, so tumors stand out as dark defects against the glowing liver, a contrast mechanism that makes the hepatobiliary phase exceptionally sensitive for cancer detection.

To evaluate the shortened protocol, the researchers assembled a retrospective cohort of 43 patients, mean age about 70 years and overwhelmingly male, who had undergone percutaneous microwave ablation for naive hepatocellular carcinoma at a tertiary care liver transplant center between January 2017 and December 2019. Nearly all had cirrhosis, most commonly from chronic hepatitis C or alcoholic liver disease. The cohort yielded 215 complete gadoxetic acid-enhanced surveillance MRI exams. Twenty-seven consecutive patients with recurrence contributed 48 positive and 86 negative exams, and 16 randomly selected patients without recurrence added another 81 negative exams, deliberately enriching the dataset with negative studies to approximate real-world screening conditions. Three board-certified abdominal radiologists, blinded to all clinical information, then independently reviewed the abbreviated versions of each exam, comparing them against the original full-protocol reports, which served as the reference standard.

The results were striking. On a per-patient basis, the abbreviated protocol achieved pooled sensitivity of 86.4 percent, specificity of 100 percent, and a negative predictive value of 88.1 percent for detecting recurrence of any type. Analyzed exam by exam, the numbers were 87.5 percent sensitivity, 97.6 percent specificity, and 93.3 percent negative predictive value, with an area under the receiver operating characteristic curve of 0.90. For recurrences confined to the ablation zone, per-patient sensitivity was 85.7 percent with perfect specificity, while for recurrences elsewhere in the liver the protocol detected 86.6 percent of cases, also with perfect specificity, though the negative predictive value for these distant recurrences dipped to 72.4 percent. Perhaps most impressive was the consistency between readers: inter-reader agreement was excellent, with a kappa of 0.91 per patient, comparable to the agreement reported for full-protocol interpretation of treated lesions.

The failures of the abbreviated scan were as informative as its successes. Across 645 readings, only three false positives occurred, attributable to motion artifact mimicking a nodule, a siderotic regenerative nodule, and a benign pseudolesion, all of which would likely be resolved by a confirmatory full MRI in clinical practice. False negatives were more numerous, eleven in total, and clustered around small recurrences measuring roughly 10 to 16 millimeters that showed arterial hyperenhancement on the complete scan but were invisible or only partially visible as defects on the hepatobiliary phase. A small minority of hepatocellular carcinomas, estimated at 5 to 10 percent, retain the transporter expression that makes them take up gadoxetic acid, the so-called green hepatocellular carcinomas, and these can evade detection on hepatobiliary-phase imaging alone.

The study builds on a growing body of evidence that abbreviated MRI works for primary screening of at-risk cirrhotic patients, where gadoxetate-enhanced abbreviated protocols have reported sensitivities in the low 80 percent range. What distinguishes the new work is its focus on secondary surveillance, the follow-up of patients already treated for hepatocellular carcinoma, a setting with far fewer published data. A Korean study by Jeon and colleagues previously found no significant difference between abbreviated and full-protocol MRI for detecting post-treatment recurrence, with per-patient sensitivity of 73.1 percent for the abbreviated approach, and the UCLA results, with higher sensitivity, reinforce the case that trimming the protocol does not meaningfully compromise cancer detection.

The clinical implications are considerable. A surveillance exam consisting of two sequences instead of the full multiphasic battery could shorten scanner time substantially, reduce per-exam cost, and expand access for the many patients who currently face months-long waits for liver MRI. Because a negative abbreviated scan reliably excludes recurrence, the authors argue, it could spare patients unnecessary comprehensive imaging, lowering both healthcare expenditure and patient burden. A positive abbreviated scan, meanwhile, would simply trigger a confirmatory complete MRI, meaning false positives carry less clinical harm than false negatives. The researchers also point to the horizon: deep learning reconstruction is already being used to denoise images and cut scan times further, and experimental synthetic contrast techniques have produced gadolinium-free contrast-enhanced MRI in as little as four minutes, compared with 28 minutes for standard protocols.

The authors are careful to note the caveats. The study was retrospective, the enriched proportion of recurrence-negative exams does not reflect a true clinical population, no formal cost or time savings were quantified, the abbreviated protocol omitted diffusion-weighted imaging, and exams spanned both 1.5 and 3 Tesla scanners. Larger prospective and multicenter trials will be needed before abbreviated protocols become standard practice. Still, for the hundreds of thousands of patients worldwide who live under the shadow of hepatocellular carcinoma recurrence, the message is hopeful: the future of surveillance may be shorter, safer, and just as sharp.

Subject of Research: Abbreviated hepatobiliary phase MRI for surveillance of hepatocellular carcinoma recurrence after microwave ablation

Article Title: Hepatobiliary phase abbreviated MRI for secondary surveillance after microwave ablation of HCC

Article References: Keshavarz, P., King, K. G., Depetris, J., Quirk, M., Kang, T. W., Sayre, J., Tong, M. J., McWilliams, J. P., Raman, S. S., & Lu, D. S. K. (2025). Hepatobiliary phase abbreviated MRI for secondary surveillance after microwave ablation of HCC. CVIR Oncology, 1(1), Article 6. https://doi.org/10.1007/s44343-025-00007-1

Image Credits: AI Generated

DOI: 10.1007/s44343-025-00007-1

Keywords: hepatocellular carcinoma, abbreviated MRI, hepatobiliary phase, microwave ablation, secondary surveillance, gadoxetic acid, liver cancer recurrence, radiology, medical imaging, cirrhosis, negative predictive value, deep learning MRI