Sprint Cycling Study


A recent study compared the effects of six 30-second maximum sprint cycling sessions with 90 minutes of continuous moderate cycling on the blood of 19 young men. The study, published in Cell Reports Medicine, August 13, measured proteins in the plasma, the liquid part of the blood, before exercise, immediately after, and three hours later.

The intense sprint cycling caused a rapid early shift in protein levels, while moderate cycling caused a smaller delayed shift. After exercise, muscles, fat tissue, liver, and immune cells change the composition of plasma, which then reaches other tissues and can alter their gene activity. The study tracked how the two cycling regimes changed the composition of plasma and its effects on fat cells.

The researchers found that 714 out of 2,884 measured proteins changed immediately after sprint cycling, while seven proteins changed immediately after moderate cycling, and 19 changed three hours later. The study also showed that the plasma taken after sprint cycling altered the activity of 1,128 genes and decreased the activity of 549 genes in human fat cells grown in the lab.

🔗 Read original →

Cell Study Reveals Role


Researchers found that depleting cells of polyamines, molecules that hold reactive iron, led to an increase in labile iron. Cells also became more dependent on GPX4, an enzyme that reduces oxidized membrane fats. Iron is essential for cells, but its reactive part triggers chain reactions that damage membrane fats, leading to ferroptosis.


The team of Ankur Jain conducted a genome-wide CRISPR screen to determine the purpose of the large polyamine reserve. They found that GPX4 was crucial, and its genetic loss or blockade with drugs was fatal to cells with depleted polyamines. Adding spermidine and ferroptosis-inhibiting substances restored cell viability.


The authors linked the polyamine reserve to membrane protection against oxidation. They investigated the cause of GPX4 dependence in iron and found that depleting polyamines increased the level of ferritin, a protein that stores iron. A fluorescent dye showed more reactive iron, and deferroxamine, an iron chelator, restored cell viability when GPX4 was blocked.

🔗 Read original →

PerturbLDM Model


The PerturbLDM model predicts how drugs change gene activity in cells under unmeasured conditions. A preprint about PerturbLDM was released on August 12 on bioRxiv. It was tested on combinations of drug, dose, and cell line that were hidden during training, but each drug, dose, and line was separately present in the data.


One and the same drug can change gene activity differently in different cell lines - populations of cells that are grown and studied separately. There are so many combinations of drug, dose, and line that laboratories can only measure a part of them. The model is needed to predict in advance which genes will increase or decrease activity in an unmeasured condition. In the preprint, PerturbLDM receives the name of the drug, dose, and cell profile of the same line in a control experiment with DMSO - a solvent that is added instead of the drug.


The model first compresses information about the activity of thousands of genes into a compact record, then gradually builds the response taking into account the drug, dose, and control profile, and then translates it back into gene activities. The authors took 46,471 conditions from Tahoe-100M - a large set of single-cell measurements. For training, they left 32,529, and 13,942 were hidden. Both parts included all 379 drugs, 47 cell lines, and three dose levels. Only a specific triplet was hidden: each of its drug, dose, and cell line was separately familiar to the model.

🔗 Read original →

Antibody Binding Model


The model, trained on antibody contact regions with targets, more accurately predicted the strength of their binding. On August 13, a study was published in Communications AI & Computing about a language model that reads sequences of both chains of an antibody. The authors hid amino acids during training, primarily in the six CDR loops, where the antibody contacts the target, and tested predictions on variants of antibodies to six antigens.


On a set of 11,052 variants, the prediction quality improved by 26.6% compared to the original model. An antibody recognizes a target with the ends of two protein chains, each with three CDR loops that form the contact surface. The rest of the chain holds the loops in the correct shape, and replacing an amino acid in a loop can change the binding strength.


The language model of proteins learns to restore hidden amino acids in a sequence. During regular training, gaps are randomly distributed throughout the chain, with some tasks falling on framework regions that are relatively similar; CDR loops are more diverse and determine which target the antibody recognizes. The Boston University team trained a model on a pair of heavy and light chains and hid 50% of amino acids within CDR loops.

🔗 Read original →

Aging Cell Type

Обри де Грей questioned whether one gene shows if a cell retains its type during partial reprogramming. On August 17, Майкл Уэст referenced a graph of two genes during brief induction of Яманаки factors - proteins that change gene function. Обри де Грей responded that an early sign of a cell transitioning to an unspecialized state grows only slightly before TERT, a gene associated with telomerase.


Their exchange raised a practical question: what changes indicate that a cell has altered its age markers but still retains its type. Яманаки factors (OSKM) lead cells to pluripotency when used long-term - a state from which different cell types can arise. With brief OSKM induction, researchers seek a regime in which age markers have shifted, but the cell retains specialization.


In a post on August 17, Уэст wrote that brief OSKM expression may "reverse developmental processes, as in cloning". For clinical application, he terms this approach "induced tissue regeneration" and references a temporal graph from a 2018 study. The graph compares ZFP42 and TERT, with ZFP42 being an early sign of movement towards pluripotency. In a response post, де Грей noted the closeness of these lines: "ZFP42 growth precedes TERT growth by only a little". A short OSKM pulse must be evaluated on a temporal map of several cell programs: age markers, signs of its original type, early and late pluripotency markers, and TERT.

🔗 Read original →

Singapore Invests in Bio-Computing


The city-state of Singapore has taken an interest in the bio-computer CL1, to the point of building data centers around these computers. Each computational unit of the CL1 contains around 200,000 neurons grown from reprogrammed stem cells. These neurons sit on a microchip with electrodes and exchange electrical impulses with the computer, converting biological activity into computational power. The bio-computers are literally fed like living organisms, with a sugar solution, microelements, and a buffer to maintain pH, as well as a gas supply system providing oxygen, carbon dioxide, and nitrogen, every three days.


The energy consumption is almost negligible, at 30 W per unit, less than that of a pocket calculator. For comparison, a server with eight Nvidia H100 chips consumes around 10,200 W. In a country where data centers already account for 7% of the total energy consumption, this is not just an innovation, but a strategic move. Currently, the Singapore center has 20 CL1 units installed, but plans to expand to 1,000 after testing safety and energy efficiency.


A similar facility in Melbourne is already operational, with 120 CL1 units and twenty clients, including universities and corporate laboratories, which pay around US$ 2,200 per month for access to the biological computing resource. This is almost half the cost of renting a top-of-the-line AI chip in the cloud. It's worth noting that these bio-computers do not compete with silicon in terms of speed and accuracy, and will not replace ChatGPT or supercomputers. Their strength lies in learning from small data and adapting to unpredictable conditions, as described in Nature Aging, July 2026.

🔗 Source: @solid_state_humanity

HistAgent Model


The HistAgent model uses a standard stained tissue section to suggest which genes to check in a tissue section. On August 18, a preprint about HistAgent was released on the Research Square platform. The authors trained the model on 2.23 million pairs of images of H&E-stained tissue sections and measured gene activity profiles in the same location.


For each section, HistAgent generates an ordered list of 50 genes. On 135 human and mouse slides not used in training, this list was compared to measurements of spatial transcriptomics, a method that shows gene activity with coordinates in the section. HistAgent considers a small section and its surroundings, then suggests an order of genes most characteristic of that point.


The authors chose the order of genes instead of continuous numerical estimates because, according to their calculations, it is less dependent on the scale of data, normalization, and differences between measurement series. To build a numerical map of gene activity across the section, a separate procedure translates these ranks into estimates based on an independent set of already measured profiles.

🔗 Read original →

New CRISPR Editor

The evoCasΦ2, a revised version of the compact CRISPR editor CasΦ2, edited DNA in human cells up to 70 times more actively than the original enzyme. On August 13, a preprint was published on bioRxiv about evoCasΦ2, a version of CasΦ2 with six amino acid substitutions. The authors compared it to the original enzyme and two versions created by rational design on four genomic sites in the human cell line HEK293T.


The CasΦ2 is a CRISPR protein that finds a target DNA site by guide RNA and cuts it. It has 757 amino acids, requiring less space in the delivery system than the larger Cas9. The original CasΦ2 barely edited DNA in mammalian cells, and successful variants were difficult to distinguish from random noise. To address this, the authors built EPICA.2, a two-step directed evolution system.


They created many slightly modified CasΦ2 variants and first selected them in yeast: DNA cutting restored gene function, allowing the cell to grow. Four rounds of mutation and re-selection gathered a library of more active variants. The library was then transferred to human cells, where a gene for green fluorescent protein was shifted and non-functional. CasΦ2 cutting sometimes restored the gene's correct reading, the protein began to glow, and the authors selected such cells. Long reading of each DNA molecule with an individual tag preserved the set of substitutions within one variant, leading to the discovery of evoCasΦ2 with six substitutions.


The effect is created by the combination of these six substitutions. On a control site, no single substitution repeated it, and removal of any from the full version reduced activity. On four native sites in the HEK293T cell genome, evoCasΦ2 edited DNA up to 70 times more actively than the original CasΦ2 and outperformed two versions created by rational design. On individual targets, the fraction of edited DNA remained below 5%. EPICA.2 continues the first EPICA, which in 2024 gave a variant of another compact nuclease, CjCas9, with activity up to 12 times higher than the original enzyme on endogenous sites in human cells, as reported in Nature Aging, July 2026.

🔗 Read original →

Cell Trainer


Researchers from Tufts University and the Wyss Institute for Biologically Inspired Engineering at Harvard University presented the open-source Cell Trainer setup on August 13, 2026. It delivers precisely defined pulses of chemicals to cells, reads their response via fluorescent molecular sensors, and adjusts the next pulse as needed.


The usual approach to managing cell culture is "bottom-up": genes are turned off, signaling pathways within the cell are altered, and constant doses of drugs are added. However, cells can adapt to such interventions by activating bypass metabolic pathways, developing resistance to the drug, or silencing the inserted gene. In March, Michael Levin noted that physiology lacks tools that account for time and feedback.


The Cell Trainer utilizes the same ability of cells to adapt: it delivers a series of stimuli and monitors how their response changes. In a preprint, Patrick Erkinson and Michael Levin's team described a complex consisting of a plate with four separate cell chambers and an automated inverted microscope. Solutions flow through thin channels in the plate, while the microscope moves underneath; the plate itself remains stationary. This setup allows for alternating observation of cells in multiple chambers without shifting the culture with a fluid flow. In a scheduled mode, the authors delivered repeated 2-minute pulses of dimethylsulfoxide to muscle precursor cells, or myoblasts, and a fluorescent sensor showed the calcium content inside each cell. From series to series, the calcium response became stronger: the repeated stimulus elicited a more pronounced reaction.


In a second experiment, the setup worked with feedback. Fibroblasts - cells of connective tissue - from a rat kidney produced a fluorescent sensor of cell membrane state and acidity. The program processed a snapshot in under 1 second; when the signal brightness exceeded a set threshold, it delivered an acidic environment for 30 seconds. This system returned the indicator to the target range. The authors released the designs of the optical block, hardware schematics, and software code, as described in Nature Aging, July 2026. The Cell Trainer provides laboratories with a reproducible way to conduct experiments on cell cultures where repeated stimuli need to be delivered, responses measured, and checked if the cell's physiological state can be directed to change.

🔗 Read original →

Heart Repair in Mice


Researchers compared genes active in the hearts of newborn mice after a heart attack and after aortic constriction. A combination of three proteins, mainly secreted by immune cells in the heart, reduced cell death and increased cell division in cellular experiments. Both effects required TLR2, a sensor protein on the surface of heart cells.


In adult hearts, heart cells are rarely renewed. In mice, there is a short window after birth when these contracting muscle cells still divide. A previous study by the same group in 2019 found that aortic constriction on the first day of life preserved heart function and was accompanied by capillary growth. If performed on the seventh day, heart function deteriorated and fibrosis occurred.


The new study sought signals that trigger early heart repair under such stress. The authors compared genes active one day after a heart attack induced on the first day of life with genes active after two weeks of aortic constriction. Among 995 common genes, they selected three proteins secreted into the surrounding tissue: CCL4, S100A8, and C1QA. These proteins have receptors on heart cells and vessel cells, and tissue staining showed that they are mainly produced by immune cells in the heart after stress.

🔗 Read original →

Cambridge to Lead £20m Center


The University of Cambridge has announced that the £20m grant from the Medical Research Council will be used to establish a new center. This center will create and distribute human tissue models grown from patient cells from the UK's National Health Service to research groups. The center's work will utilize artificial intelligence methods. Before testing a drug on humans, developers need to understand how the substance affects the affected tissue. A model grown from a patient's cells retains some of the patient's biology and allows observation of the reaction in human cells.


One such tool is an organoid, a small, simplified tissue model grown from a patient's cells. Such models are made for the intestine, tumors, and brain. The new center will provide these models to universities and pharmaceutical companies for early candidate testing. Center director Matthias Zilbauer explains: "Since these miniature human tissues retain many unique biological features of a specific patient, they allow for more accurate study of the disease and testing of potential treatments before clinical trials."


The center aims to develop models that can be reproduced and applied in different laboratories. This way, results from patient tissues can participate in early drug candidate selection beyond a single laboratory. In November 2025, the UK government published the Replacing Animals in Science strategy and allocated £75m for it. The new center adds to this program by providing shared human tissue models for drug development.

🔗 Read original →

Early Menopause in Mice


Researchers found that female mice born after a full IVF route had a reduced ovarian reserve. A study published in the Journal of Clinical Investigation on August 17 examined female mice born after undergoing hormonal stimulation, in vitro fertilization, embryo culture, and transfer to a recipient female. At 12 and 39 weeks, these mice had fewer follicles in their ovaries and lower estrogen levels compared to mice conceived naturally.

These female mice also had fewer live births in four consecutive pregnancies. The ovary stores immature eggs in follicles, structures where the egg grows and receives support. The authors investigated whether the full IVF route affects the future function of the ovary. They found that the number of germ cells in the ovaries of both groups was the same at the end of fetal development.

However, at 12 and 39 weeks, the authors counted fewer follicles in the IVF group, and their ovaries were smaller relative to body mass, with lower estrogen levels in the blood. The authors linked these measurements to a faster depletion of the follicular reserve after birth. Changes were also observed in subsequent pregnancies, with a higher frequency of resorbed embryos in the uterus and fewer live births in a series of four pregnancies.

🔗 Read original →

Cell Model PIANO


The PIANO model is designed to distinguish differences between individual cells from differences between donors and instruments. On August 12, in a preprint on bioRxiv, authors described PIANO, a model that integrates data on gene activity in individual cells.


The model was trained on the Tahoe-100M atlas with over 100 million cancer cell profiles in 21.8 hours on a single NVIDIA A100 graphics processor. Single-cell RNA sequencing shows which genes are active in each cell, and these measurements are used to build tissue, tumor, and brain atlases. However, a single dataset may contain cells from different people, laboratories, and instruments, which can be mistaken for cell differences when compared directly.


PIANO compresses the activity of thousands of genes into a short numerical record. When restoring the gene activity profile, the model is separately informed of the experimental conditions: donor, species, measurement platform, or drug dose. The model can then attribute part of the differences to the experimental conditions and use the record for cell characteristics. Another network tries to guess the laboratory batch - a group of cells processed together in one run - from the record.


During training, PIANO weakens this connection. The authors checked if this approach preserves different cell types separately, by removing a subset of neurons - eSPN - from non-human primate datasets. In the human dataset, eSPN remained, and PIANO preserved it separately from related neurons common to all primates, as described in Nature Aging, July 2026. On Tahoe-100M, the authors checked if the model can reproduce changes in gene activity at a given drug dose, using trametinib, which blocks the MEK protein in a signaling pathway that controls cell growth and division.

🔗 Read original →

Ebola Outbreak


The World Health Organization (WHO) is supporting the PARTNERS trial, which compares treatment options for Bundibugyo virus disease without discontinuing standard medical care. A separate team at the University of Oxford has begun an early trial of a candidate vaccine in volunteers. By August 17, epidemiologist Devi Sridhar wrote in a column for The Guardian that the outbreak in the Democratic Republic of Congo had reached an estimated 4,900 cases and over 2,300 deaths.


The WHO-backed PARTNERS trial randomly assigns patients with confirmed Bundibugyo disease to different antiviral treatment options. In the Democratic Republic of Congo, it is coordinated by the National Institute of Biomedical Research (INRB), a research center of the country's Ministry of Health. All participants receive standard medical care and are monitored for at least 28 days, allowing researchers to compare which treatment regimen is more effective in helping patients survive.


The vaccine is being tested separately: on July 24, the first volunteer at the University of Oxford received the ChAdOx1 BDBV candidate vaccine against Bundibugyo. In the first phase, the team is checking its safety and immune response in volunteers, while the treatment is being studied in patients in the outbreak zone, as reported in The Guardian, August 2023. The possibility of including a patient in the trial depends on how quickly they reach medical help, according to Nature Aging, July 2026.

🔗 Read original →

Gromov Hypothesis Solved

The Gromov hypothesis on the growth of space volume has been claimed to be solved in three mathematical manuscripts. Two of these manuscripts reveal the role of AI in the proofs, which were published on arXiv on 13-14 August. In two of the manuscripts, the authors describe the role of generative models in finding the proof.

In 1986, Mikhail Gromov asked if the geometry at each point could limit the growth of an infinite space as a whole. He considered spaces where two mathematical indicators of how space is curved around a point satisfy strict conditions: one is nowhere negative, the other is everywhere positive. For a sphere of radius R in n dimensions, Gromov expected an upper bound of the volume to be C·Rⁿ⁻². Usually, the volume in n dimensions grows like Rⁿ; here the power of the radius is less by two.

The works of Jian Ge, Joacchino Antonelli, and Bochao Kong and Xinyu Zhu all claim this bound. The path from local curvature to volume goes through the spread of heat in space in Ge's work. Antonelli obtains this estimate as a special case of a more general theorem on intermediate curvature. Kong and Zhu relate mass transfer to the volume of spheres of large radius. The three works arrive at the same estimate by different methods.

In two of the manuscripts, the authors separately describe the role of generative models in finding the proofs, citing arXiv and referencing their own work, as well as the assistance of GPT and ChatGPT 5.6 Sol Ultra and Codex in exploring the proof and writing the manuscript.

🔗 Read original →

Gene Needed for Ear Cell Regeneration


Researchers studied the regeneration of sensory cells in the inner ear of zebrafish larvae. They created a model of inner ear damage and tracked the origin of new hair cells, investigating the role of the dlx5a gene in their regeneration. The vestibular system, which includes hair cells in the inner ear, helps maintain balance by converting head movements into nerve signals.


In the study published in Nature Communications on August 18, the authors introduced the antibiotic neomycin into the inner ear of five-day-old zebrafish larvae, resulting in a 60% reduction of hair cells in the semicircular canals. New cells began to appear within 12 hours, and the vestibulo-ocular reflex returned to control levels within 48 hours.


The authors then tracked the origin of the new cells, labeling supporting cells, which are epithelial cells neighboring hair cells. Some labeled cells acquired hair cell characteristics without visible division, while others divided first. Another label showed that some restored hair cells originated from supporting cells that appeared after the injury, indicating that new hair cells arise through multiple pathways.


To investigate changes in gene function during regeneration, the authors measured gene activity in individual inner ear cells at 6, 24, and 48 hours after damage. They found the most changes in gene activity at 24 hours, with the dlx5a gene, which encodes a protein regulating other genes, being highlighted in supporting cells. The authors used the CRISPR–Cas9 system to disable dlx5a in supporting cells, resulting in poorer hair cell regeneration after injury.

🔗 Read original →

FoxM1 Protein Study


Researchers at Kyoto University compared two studies on the FoxM1 protein, which regulates cell division and DNA repair programs. In one study, FoxM1 was briefly activated in young mice, while in the other, it was suppressed in very old mice, removing some senescent cells. The authors linked the difference to age, organ, and cell type.


The review, published on August 17, compared these results, noting that senescent cells stop dividing but remain in tissue, releasing inflammatory signals. FoxM1 may support their survival, and when briefly activated from a young age, tissue restoration programs. In a 2022 study, researchers activated an added active version of FoxM1 in mice from eight weeks of age: three days of work, four days of pause, for a total of 80 weeks.


The authors reported that in progeroid and naturally aging animals, age-related signs weakened, and lifespan increased. In a 2025 article, the authors investigated the link between PGAM1 and Chk1 in senescent cells, finding that their interaction stabilized HIF-2α, a regulator of glucose metabolism genes, and increased FoxM1 activity. The cell received material for nucleotide synthesis, the building blocks of DNA, and maintained a survival program, as described in the Nature Aging, July 2026.

🔗 Read original →

Anthropic Success


Anthropic designed 1,320 miniproteins for 15 targets, and two laboratories verified their binding on August 18. Anthropic published the campaign results, a full technical report, and open data. Out of 1,320 sequences, 354 were recognized as binding by the combined assessment of two contract laboratories.

The miniprotein in this campaign was required to attach to a selected protein target. To achieve this, a site on the surface of the target needed to be chosen, and the shape and amino acid sequence of the candidate had to be selected, manufactured, and measured for binding. Anthropic selected the targets and prepared a general protocol. Claude studied each target, chose a site on its surface, launched open programs to build the shape and select amino acids, checked candidates on a computer, and selected the final lists for the laboratory.

In Nature Aging, July 2026, similar research was conducted. The current campaign involved a longer chain: the agent itself chose what and how to design, and the laboratory measured each final candidate on a series of targets. In May, Adaptyv Bio tested 100 protein candidates created by ten human teams and six AI agents for TREM2, a protein on immune cells in the brain. In the current campaign, this laboratory participated in verifying 15 targets, and Anthropic revealed the results for each transmitted sequence. Anthropic passed the final sequences to two contract laboratories - Adaptyv Bio and Twist Bioscience.

🔗 Read original →

Aging Cells Impact Therapy


The journal Frontiers in Aging published a review on August 17, 2026, about mesenchymal stromal cells (MSC) and their use in tissue repair and inflammation management. The authors explain that basic quality control may not be enough, as the product may appear viable but perform worse in specific therapies. The review outlines the path of the cell product from donor to patient, including growth, freezing, transportation, and thawing, where cells may experience stress at each stage.

Cell aging is a persistent state where cells divide less often, change their metabolism, and alter the signals they send to surrounding tissue, leading to a decrease in the product's ability to perform its intended function. The authors note that the product may retain acceptable viability and surface markers but have already lost important functionality. The review links stress to loss of function through a chain of events, including DNA damage and mitochondrial dysfunction.

The source of the cells determines where this chain begins, with bone marrow-derived MSC being affected by donor age and inflammation in the bone marrow. Cells from umbilical cord tissue are typically younger but are affected by growth, freezing, thawing, and transportation. For adipose tissue-derived MSC, donor obesity, diabetes, and metabolic inflammation are important factors. The authors suggest linking testing to the product's intended use, such as checking if cells can suppress T-cell activation for inflammation or observing vascular cell migration for tissue repair.

🔗 Read original →

NeuroQuant PET Approved


The FDA has approved NeuroQuant PET, a program that translates amyloid PET scans into a common scale. On August 12, the FDA granted a "Substantially Equivalent" decision for NeuroQuant PET under the K261916 application. The program automatically generates a quantitative report for amyloid PET scans in patients with cognitive impairments and other neurological conditions.


The amyloid PET scan shows where a tracer accumulates in the brain, binding to amyloid deposits. However, the resulting number depends on the tracer, scan time, device, and calculation method, making it difficult to compare results from different studies and clinics. The Centiloid scale translates these results into common units, with zero corresponding to the average value in young individuals with high confidence in the absence of amyloid, and 100 corresponding to the average value in typical Alzheimer's disease patients.


NeuroQuant PET performs this translation for a given scan, highlighting brain regions, calculating SUVr, and adding the Centiloid value to a structured report. According to Cortechs.ai, the FDA approval applies to amyloid reports for flutemetamol, florbetaben, and florbetapir. The scale, created for research comparison, becomes part of the automatic clinical report, supplementing the image with a number that is equally readable across different tracers and analysis methods.

🔗 Read original →

NeuroQuant PET Approved


The FDA has approved NeuroQuant PET, a program that translates amyloid PET scans into a common scale. On August 12, the FDA granted a "Substantially Equivalent" decision for NeuroQuant PET under the K261916 application. The program automatically generates a quantitative report for amyloid PET scans in patients with cognitive impairments and other neurological conditions.


The amyloid PET scan shows where a tracer accumulates in the brain, binding to amyloid deposits. However, the resulting number depends on the tracer, scan time, device, and calculation method, making it difficult to compare results from different studies and clinics. The Centiloid scale translates these results into common units, with 0 corresponding to the average value in young individuals with high confidence in the absence of amyloid, and 100 corresponding to the average value in typical Alzheimer's disease patients.


NeuroQuant PET performs this translation for a given scan, highlighting brain regions, calculating SUVr, and adding the Centiloid value to a structured report. According to Cortechs.ai, the FDA approval extends to amyloid reports for flutemetamol, florbetaben, and florbetapir. The scale, created for research comparison, becomes part of the automatic clinical report, supplementing the image with a number that is equally readable across different tracers and analysis methods.

🔗 Read original →