The shape and structure of blood cells provide vital indicators for diagnosis and management of blood disease and disorders. Recognizing subtle differences in the appearance of cells under a microscope, however, requires the skills of experts with years of training, motivating researchers to investigate whether artificial intelligence (AI) could help automate this onerous task. A UK-led research team has now developed a generative AI-based model, known as CytoDiffusion, that characterizes blood cell morphology with greater accuracy and reliability than human experts.

Conventional discriminative machine learning models can match human performance at classifying cells in blood samples into predefined classes. But discriminative models, which learn to recognise cell images based on expert labels, struggle with never-before-seen cell types and images from differing microscopes and staining techniques.

To address these shortfalls, the team – headed up at the University of Cambridge, University College London and Queen Mary University of London – created CytoDiffusion around a diffusion-based generative AI classifier. Rather than just learning to separate cell categories, CytoDiffusion models the full range of blood cell morphologies to provide accurate classification with robust anomaly detection.

“Our approach is motivated by the desire to achieve a model with superhuman fidelity, flexibility and metacognitive awareness that can capture the distribution of all possible morphological appearances,” the researchers write.

Authenticity and accuracy

For AI-based analysis to be adopted in the clinic, it’s essential that users trust a model’s learned representations. To assess whether CytoDiffusion could effectively capture the distribution of blood cell images, the team used it to generate synthetic blood cell images. Analysis by experienced haematologists revealed that these synthetic images were near-indistinguishable from genuine images, showing that CytoDiffusion genuinely learns the morphological distribution of blood cells rather than using artefactual shortcuts.

The researchers used multiple datasets to develop and evaluate their diffusion classifier, including CytoData, a custom dataset containing more than half a million anonymized cell images from almost 3000 blood smear slides. In standard classification tasks across these datasets, CytoDiffusion achieved state-of-the-art performance, matching or exceeding the capabilities of traditional discriminative models.

Effective diagnosis from blood smear samples also requires the ability to detect rare or previously unseen cell types. The researchers evaluated CytoDiffusion’s ability to detect blast cells (immature blood cells) in the test datasets. Blast cells are associated with blood malignancies such as leukaemia, and high detection sensitivity is essential to minimize false negatives.

In one dataset, CytoDiffusion detected blast cells with sensitivity and specificity of 0.905 and 0.962, respectively. In contrast, a discriminative model exhibited a poor sensitivity of 0.281. In datasets with erythroblasts as the abnormal cells, CytoDiffusion again outperformed the discriminative model, demonstrating that it can detect abnormal cell types not present in its training data, with the high sensitivity required for clinical applications.

Robust model

It’s important that a classification model is robust to different imaging conditions and can function with sparse training data, as commonly found in clinical applications. When trained and tested on diverse image datasets (different hospitals, microscopes and staining procedures), CytoDiffusion achieved state-of-the-art accuracy in all cases. Likewise, after training on limited subsets of 10, 20 and 50 images per class, CytoDiffusion consistently outperformed discriminative models, particularly in the most data-scarce conditions.

Another essential feature of clinical classification tasks, whether performed by a human or an algorithm, is knowing the uncertainty in the final decision. The researchers developed a framework for evaluating uncertainty and showed that CytoDiffusion produced superior uncertainty estimates to human experts. With uncertainty quantified, cases with high certainty could be processed automatically, with uncertain cases flagged for human review.

“When we tested its accuracy, the system was slightly better than humans,” says first author Simon Deltadahl from the University of Cambridge in a press statement. “But where it really stood out was in knowing when it was uncertain. Our model would never say it was certain and then be wrong, but that is something that humans sometimes do.”

Finally, the team demonstrated CytoDiffusion’s ability to create heat maps highlighting regions that would need to change for an image to be reclassified. This feature provides insight into the model’s decision-making process and shows that it understands subtle differences between similar cell types. Such transparency is essential for clinical deployment of AI, making models more trustworthy as practitioners can verify that classifications are based on legitimate morphological features.

“The true value of healthcare AI lies not in approximating human expertise at lower cost, but in enabling greater diagnostic, prognostic and prescriptive power than either experts or simple statistical models can achieve,” adds co-senior author Parashkev Nachev from University College London.

CytoDiffusion is described in Nature Machine Intelligence.

The post Generative AI model detects blood cell abnormalities appeared first on Physics World.

Structural colours – created using nanostructures that scatter and reflect specific wavelengths of light – offer a non-toxic, fade-resistant and environmentally friendly alternative to chemical dyes. Large-scale production of structural colour-based materials, however, has been hindered by fabrication challenges and a lack of effective tuning mechanisms.

In a step towards commercial viability, a team at the University of Central Florida has used vanadium dioxide (VO₂) – a material with temperature-sensitive optical and structural properties – to generate tunable structural colour on both rigid and flexible surfaces, without requiring complex nanofabrication.

Senior author Debashis Chanda and colleagues created their structural colour platform by stacking a thin layer of VO₂ on top of a thick, reflective layer of aluminium to form a tunable thin-film cavity. At specific combinations of VO₂ grain size and layer thickness this structure strongly absorbs certain frequency bands of visible light, producing the appearance of vivid colours.

The key enabler of this approach is the fact that at a critical transition temperature, VO₂ reversibly switches from insulator to metal, accompanied by a change in its crystalline structure. This phase change alters the interference conditions in the thin-film cavity, varying the reflectance spectra and changing the perceived colour. Controlling the thickness of the VO₂ layer enables the generation of a wide range of structural colours.

The bilayer structures are grown via a combination of magnetron sputtering and electron-beam deposition, techniques compatible with large-scale production. By adjusting the growth parameters during fabrication, the researchers could broaden the colour palette and control the temperature at which the phase transition occurs. To expand the available colour range further, they added a third ultrathin layer of high-refractive index titanium dioxide on top of the bilayer.

The researchers describe a range of applications for their flexible coloration platform, including a colour-tunable maple leaf pattern, a thermal sensing label on a coffee cup and tunable structural coloration on flexible fabrics. They also demonstrated its use on complex shapes, such as a toy gecko with a flexible tunable colour coating and an embedded heater.

“These preliminary demonstrations validate the feasibility of developing thermally responsive sensors, reconfigurable displays and dynamic colouration devices, paving the way for innovative solutions across fields such as wearable electronic, cosmetics, smart textiles and defence technologies,” the team concludes.

The research is described in Proceedings of the National Academy of Sciences.

The post Phase-changing material generates vivid tunable colours appeared first on Physics World.

Positron emission tomography (PET) is used extensively within preclinical research, enabling molecular imaging of rodent brains, for example, to investigate neurodegenerative disease. Such imaging studies require the highest possible spatial resolution to resolve the tiny structures in the animal’s brain. A research team at the National Institutes for Quantum Science and Technology (QST) in Japan has now developed the first PET scanner to achieve sub-0.5 mm spatial resolution.

Submillimetre-resolution PET has been demonstrated by several research groups. Indeed, the QST team previously built a PET scanner with 0.55 mm resolution – sufficient to visualize the thalamus and hypothalamus in the mouse brain. But identification of smaller structures such as the amygdala and cerebellar nuclei has remained a challenge.

“Sub-0.5 mm resolution is important to visualize mouse brain structures with high quantification accuracy,” explains first author Han Gyu Kang. “Moreover, this research work will change our perspective about the fundamental limit of PET resolution, which had been regarded to be around 0.5 mm due to the positron range of [the radioisotope] fluorine-18”.

System optimization

With Monte Carlo simulations revealing that sub-0.5 mm resolution could be achievable with optimal detector parameters and system geometry, Kang and colleagues performed a series of modifications to their submillimetre-resolution PET (SR-PET) to create the new high-resolution PET (HR-PET) scanner.

The HR-PET, described in IEEE Transactions on Medical Imaging, is based around two 48 mm-diameter detector rings with an axial coverage of 23.4 mm. Each ring contains 16 depth-of-interaction (DOI) detectors (essential to minimize parallax error in a small ring diameter) made from three layers of LYSO crystal arrays stacked in a staggered configuration, with the outer layer coupled to a silicon photomultiplier (SiPM) array.

Compared with their previous design, the researchers reduced the detector ring diameter from 52.5 to 48 mm, which served to improve geometrical efficiency and minimize the noncollinearity effect. They also reduced the crystal pitch from 1.0 to 0.8 mm and the SiPM pitch from 3.2 to 2.4 mm, improving the spatial resolution and crystal decoding accuracy, respectively.

Other changes included optimizing the crystal thicknesses to 3, 3 and 5 mm for the first, second and third arrays, as well as use of a narrow energy window (440–560 keV) to reduce the scatter fraction and inter-crystal scattering events. “The optimized staggered three-layer crystal array design is also a key factor to enhance the spatial resolution by improving the spatial sampling accuracy and DOI resolution compared with the previous SR-PET,” Kang points out.

Performance tests showed that the HR-PET scanner had a system-level energy resolution of 18.6% and a coincidence timing resolution of 8.5 ns. Imaging a NEMA ²²Na point source revealed a peak sensitivity at the axial centre of 0.65% for the 440–560 keV energy window and a radial resolution of 0.67±0.06 mm from the centre to 10 mm radial offset (using 2D filtered-back-projection reconstruction) – a 33% improvement over that achieved by the SR-PET.

To further evaluate the performance of the HR-PET, the researchers imaged a rod-based resolution phantom. Images reconstructed using a 3D ordered-subset-expectation-maximization (OSEM) algorithm clearly resolved all of the rods. This included the smallest rods with diameters of 0.5 and 0.45 mm, with average valley-to-peak ratios of 0.533 and 0.655, respectively – a 40% improvement over the SR-PET.

In vivo brain PET

The researchers then used the HR-PET for in vivo mouse brain imaging. They injected ¹⁸F-FITM, a tracer used to image the central nervous system, into an awake mouse and performed a 30 min PET scan (with the animal anesthetized) 42 min after injection. For comparison, they scanned the same mouse for 30 min with a preclinical Inveon PET scanner.

After OSEM reconstruction, strong tracer uptake in the thalamus, hypothalamus, cerebellar cortex and cerebellar nuclei was clearly visible in the coronal HR-PET images. A zoomed image distinguished the cerebellar nuclei and flocculus, while sagittal and axial images visualized the cortex and striatum. Images from the Inveon, however, could barely resolve these brain structures.

The team also imaged the animal’s glucose metabolism using the tracer ¹⁸F-FDG. A 30 min HR-PET scan clearly delineated glucose transporter expression in the cortex, thalamus, hypothalamus and cerebellar nuclei. Here again, the Inveon could hardly identify these small structures.

The researchers note that the ¹⁸F-FITM and ¹⁸F-FDG PET images matched well with the anatomy seen in a preclinical CT scan. “To the best of our knowledge, this is the first separate identification of the hypothalamus, amygdala and cerebellar nuclei of mouse brain,” they write.

Future plans for the HR-PET scanner, says Kang, include using it for research on neurodegenerative disorders, with tracers that bind to amyloid beta or tau protein. “In addition, we plan to extend the axial coverage over 50 mm to explore the whole body of mice with sub-0.5 mm resolution, especially for oncological research,” he says. “Finally, we would like to achieve sub-0.3 mm PET resolution with more optimized PET detector and system designs.”

The post High-resolution PET scanner visualizes mouse brain structures with unprecedented detail appeared first on Physics World.

Artificial muscles that offer flexible functionality could prove invaluable for a range of applications, from soft robotics and wearables to biomedical instrumentation and minimally invasive surgery. Current designs, however, are limited by complex actuation mechanisms and challenges in miniaturization. Aiming to overcome these obstacles, a research team headed up at the Acoustic Robotics Systems Lab (ETH Zürich) in Switzerland is using microbubbles to create soft, programmable artificial muscles that can be wirelessly controlled via targeted ultrasound activation.

Gas-filled microbubbles can concentrate acoustic energy, providing a means to initiate movement with rapid response times and high spatial accuracy. In this study, reported in Nature, team leader Daniel Ahmed and colleagues built a synthetic muscle from a thin flexible membrane containing arrays of more than 10,000 microbubbles. When acoustically activated, the microbubbles generate thrust and cause the membrane to deform. And as different sized microbubbles resonate at different ultrasound frequencies, the arrays can be designed to provide programmable motion.

“Ultrasound is safe, non-invasive, can penetrate deep into the body and can generate large forces. However, without microbubbles, a much higher force is needed to deform the muscle, and selective activation is difficult,” Ahmed explains. “To overcome this limitation, we use microbubbles, which amplify force generation at specific sites and act as resonant systems. As a result, we can activate the artificial muscle at safe ultrasound power levels and generate complex motion.”

The team created the artificial muscles from a thin silicone membrane patterned with an array of cylindrical microcavities with the dimensions of the desired microbubbles. Submerging this membrane in a water-filled acoustic chamber trapped tens of thousands of gas bubbles within the cavities (one per cavity). The final device contains around 3000 microbubbles per mm² and weighs just 0.047 mg/mm².

To demonstrate acoustic activation, the researchers fabricated an artificial muscle containing uniform-sized microbubbles on one surface. They fixed one end of the muscle and exposed it to resonant frequency ultrasound, simultaneously exciting the entire microbubble array. The resulting oscillations generated acoustic streaming and radiation forces, causing the muscle to flex upward, with an amplitude dependent upon the ultrasound excitation voltage.

Next, the team designed an 80 µm-thick, 3 x 0.5 cm artificial muscle containing arrays of three different sized microbubbles. Stimulation at 96.5, 82.3 and 33.2 kHz induced deformations in regions containing bubbles with diameters of 12, 16 and 66 µm, respectively. Exposure to swept-frequency ultrasound covering the three resonant frequencies sequentially activated the different arrays, resulting in an undulatory motion.

A multitude of functions

Ahmed and colleagues showcased a range of applications for the artificial muscle by integrating microbubble arrays into functional devices, such as a miniature soft gripper for trapping and manipulating fragile live animals. The gripper comprises six to ten microbubble array-based “tentacles” that, when subjected to ultrasound, gently gripped a zebrafish larva with sub-100 ms response time. When the ultrasound was switched off, the tentacles opened and the larva swam away with no adverse effects.

The artificial muscle can function as a conformable robotic skin that sticks and imparts motion to a stationary object, which the team demonstrated by attaching it to the surface of an excised pig heart. It can also be employed for targeted drug delivery – shown by the use of a microbubble-array robotic patch for ultrasound-enhanced delivery of dye into an agar block.

The researchers also built an ultrasound-powered “stingraybot”, a soft surgical robot with artificial muscles (arrays of differently sized microbubbles) on either side to mimic the pectoral fins of a stingray. Exposure to swept-frequency ultrasound induced an undulatory motion that wirelessly propelled the 4 cm-long robot forward at a speed of about 0.8 body lengths per second.

To demonstrate future practical biomedical applications, such as supporting minimally invasive surgery or site-specific drug release within the gastrointestinal tract, the researchers encapsulated a rolled up stingraybot within a 27 x 12 mm edible capsule. Once released into the stomach, the robot could be propelled on demand under ultrasound actuation. They also pre-folded a linear artificial muscle into a wheel shape and showed that swept ultrasound frequencies could propel it along the complex mucosal surfaces of the stomach and intestine.

“Through the strategic use of microbubble configurations and voltage and frequency as ultrasound excitation parameters, we engineered a diverse range of preprogrammed movements and demonstrated their applicability across various robotic platforms,” the researchers write. “Looking ahead, these artificial muscles hold transformative potential across cutting-edge fields such as soft robotics, haptic medical devices and minimally invasive surgery.”

Ahmed says that the team is currently developing soft patches that can conform to biological surfaces for drug delivery inside the bladder. “We are also designing soft, flexible robots that can wrap around a tumour and release drugs directly at the target site,” he tells Physics World. “Basically, we’re creating mobile conformable drug-delivery patches.”

The post Microbubbles power soft, programmable artificial muscles appeared first on Physics World.

Microcirculation – the flow of blood through the smallest vessels – is responsible for distributing oxygen and nutrients to tissues and organs throughout the body. Mapping this flow at the whole-organ scale could enhance our understanding of the circulatory system and improve diagnosis of vascular disorders. With this aim, researchers at the Institute Physics for Medicine Paris (Inserm, ESPCI-PSL, CNRS) have combined 3D ultrasound localization microscopy (ULM) with a multi-lens array method to image blood flow dynamics in entire organs with micrometric resolution, reporting their findings in Nature Communications.

“Beyond understanding how an organ functions across different spatial scales, imaging the vasculature of an entire organ reveals the spatial relationships between macro- and micro-vascular networks, providing a comprehensive assessment of its structural and functional organization,” explains senior author Clement Papadacci.

The 3D ULM technique works by localizing intravenously injected microbubbles. Offering a spatial resolution roughly ten times finer than conventional ultrasound, 3D ULM can map and quantify micro-scale vascular structures. But while the method has proved valuable for mapping whole organs in small animals, visualizing entire organs in large animals or humans is hindered by the limitations of existing technology.

To enable wide field-of-view coverage while maintaining high-resolution imaging, the team – led by PhD student Nabil Haidour under Papadacci’s supervision – developed a multi-lens array probe. The probe comprises an array of 252 large (4.5 mm²) ultrasound transducer elements. The use of large elements increases the probe’s sensitive area to a total footprint of 104 x 82 mm, while maintaining a relatively low element count.

Each transducer element is equipped with an individual acoustic diverging lens. “Large elements alone are too directive to create an image, as they cannot generate sufficient overlap or interference between beams,” Papadacci explains. “The acoustic lenses reduce this directivity, allowing the elements to focus and coherently combine signals in reception, thus enabling volumetric image formation.”

Whole-organ imaging

After validating their method via numerical simulations and phantom experiments, the team used a multi-lens array probe driven by a clinical ultrasound system to perform 3D dynamic ULM of an entire explanted porcine heart – considered an ideal cardiac model as its vascular anatomies and dimensions are comparable to those of humans.

The heart was perfused with microbubble solution, enabling the probe to visualize the whole coronary microcirculation network over a large volume of 120 x 100 x 82 mm, with a spatial resolution of around 125 µm. The technique enabled visualization of both large vessels and the finest microcirculation in real time. The team also used a skeletonization algorithm to measure vessel radii at each voxel, which ranged from approximately 75 to 600 µm.

As well as structural imaging, the probe can also assess flow dynamics across all vascular scales, with a high temporal resolution of 312 frames/s. By tracking the microbubbles, the researchers estimated absolute flow velocities ranging from 10 mm/s in small vessels to over 300 mm/s in the largest. They could also differentiate arteries and veins based on the flow direction in the coronary network.

In vivo demonstrations

Next, the researchers used the multi-lens array probe to image the entire kidney and liver of an anaesthetized pig at the Veterinary school of Maison Alfort, with the probe positioned in front of the kidney or liver, respectively, and held using an articulated arm. They employed electrocardiography to synchronize the ultrasound acquisitions with periods of minimal respiratory motion and injected microbubble solution intravenously into the animal’s ear.

The probe mapped the vascular network of the kidney over a 60 x 80 x 40 mm volume with a spatial resolution of 147 µm. The maximum 3D absolute flow velocity was approximately 280 mm/s in the large vessels and the vessel radii ranged from 70 to 400 µm. The team also used directional flow measurements to identify the arterial and venous flow systems.

Liver imaging is more challenging due to respiratory, cardiac and stomach motions. Nevertheless, 3D dynamic ULM enabled high-depth visualization of a large volume of liver vasculature (65 x 100 x 82 mm) with a spatial resolution of 200 µm. Here, the researchers used dynamic velocity measurement to identify the liver’s three blood networks (arterial, venous and portal veins).

“The combination of whole-organ volumetric imaging with high-resolution vascular quantification effectively addresses key limitations of existing modalities, such as ultrasound Doppler imaging, CT angiography and 4D flow MRI,” they write.

Clinical applications of 3D dynamic ULM still need to be demonstrated, but Papadacci suggests that the technique has strong potential for evaluating kidney transplants, coronary microcirculation disorders, stroke, aneurysms and neoangiogenesis in cancer. “It could also become a powerful tool for monitoring treatment response and vascular remodelling over time,” he adds.

Papadacci and colleagues anticipate that translation to human applications will be possible in the near future and plan to begin a clinical trial early in 2026.

The post Ultrasound probe maps real-time blood flow across entire organs appeared first on Physics World.

A research team headed up at Wayne State University School of Medicine in the US has developed a novel treatment for glioblastoma, based on exposure to low levels of radiofrequency electromagnetic fields (RF EMF). The researchers demonstrated that the new therapy slows the growth of glioblastoma cells in vitro and, for the first time, showed its feasibility and clinical impact in patients with brain tumours.

The study, led by Hugo Jimenez and reported in Oncotarget, uses a device developed by TheraBionic that delivers amplitude-modulated 27.12 MHz RF EMF throughout the entire body, via a spoon-shaped antenna placed on the tongue. Using tumour-specific modulation frequencies, the device has already received US FDA approval for treating patients with advanced hepatocellular carcinoma (HCC, a liver cancer), while its safety and effectiveness are currently being assessed in clinical trials in patients with pancreatic, colorectal and breast cancer.

In this latest work, the team investigated its use in glioblastoma, an aggressive and difficult to treat brain tumour.

To identify the particular frequencies needed to treat glioblastoma, the team used a non-invasive biofeedback method developed previously to study patients with various types of cancer. The process involves measuring variations in skin electrical resistance, pulse amplitude and blood pressure while individuals are exposed to low levels of amplitude-modulated frequencies. The approach can identify the frequencies, usually between 1 Hz and 100 kHz, specific to a single tumour type.

Jimenez and colleagues first examined the impact of glioblastoma-specific amplitude-modulated RF EMF (GBMF) on glioblastoma cells, exposing various cell lines to GBMF for 3 h per day at the exposure level used for patient treatments. After one week, GBMF decreased the proliferation of three glioblastoma cell lines (U251, BTCOE-4765 and BTCOE-4795) by 34.19%, 15.03% and 14.52%, respectively.

The team note that the level of this inhibitive effect (15–34%) is similar to that observed in HCC cell lines (19–47%) and breast cancer cell lines (10–20%) treated with tumour-specific frequencies. A fourth glioblastoma cell line (BTCOE-4536) was not inhibited by GBMF, for reasons currently unknown.

Next, the researchers examined the effect of GBMF on cancer stem cells, which are responsible for treatment resistance and cancer recurrence. The treatment decreased the tumour sphere-forming ability of U251 and BTCOE-4795 cells by 36.16% and 30.16%, respectively – also a comparable range to that seen in HCC and breast cancer cells.

Notably, these effects were only induced by frequencies associated with glioblastoma. Exposing glioblastoma cells to HCC-specific modulation frequencies had no measurable impact and was indistinguishable from sham exposure.

Looking into the underlying treatment mechanisms, the researchers hypothesized that – as seen in breast cancer and HCC – glioblastoma cell proliferation is mediated by T-type voltage-gated calcium channels (VGCC). In the presence of a VGCC blocker, GBMF did not inhibit cell proliferation, confirming that GBMF inhibition of cell proliferation depends on T-type VGCCs, in particular, a calcium channel known as CACNA1H.

The team also found that GBMF blocks the growth of glioblastoma cells by modulating the “Mitotic Roles of Polo-Like Kinase” signalling pathway, leading to disruption of the cells’ mitotic spindles, critical structures in cell replication.

A clinical first

Finally, the researchers used the TheraBionic device to treat two patients: a 38-year-old patient with recurrent glioblastoma and a 47-year-old patient with the rare brain tumour oligodendroglioma. The first patient showed signs of clinical and radiological benefit following treatment; the second exhibited stable disease and tolerated the treatment well.

“This is the first report showing feasibility and clinical activity in patients with brain tumour,” the authors write. “Similarly to what has been observed in patients with breast cancer and hepatocellular carcinoma, this report shows feasibility of this treatment approach in patients with malignant glioma and provides evidence of anticancer activity in one of them.”

The researchers add that a previous dosimetric analysis of this technique measured a whole-body specific absorption rate (SAR, the rate of energy absorbed by the body when exposed to RF EMF) of 1.35 mW/kg and a peak spatial SAR (over 1 g of tissue) of 146–352 mW/kg. These values are well within the safety limits set by the ICNIRP (whole-body SAR of 80 mW/kg; peak spatial SAR of 2000 mW/kg). Organ-specific values for grey matter, white matter and the midbrain also had mean SAR ranges well within the safety limits.

The team concludes that the results justify future preclinical and clinical studies of the TheraBionic device in this patient population. “We are currently in the process of designing clinical studies in patients with brain tumors,” Jimenez tells Physics World.

The post Tumour-specific radiofrequency fields suppress brain cancer growth appeared first on Physics World.

Pancreatic ductal adenocarcinoma (PDAC), the most common type of pancreatic cancer, is an aggressive tumour with a poor prognosis. Surgery remains the only potential cure, but is feasible in just 10–15% of cases. A team headed up at Sichuan University in China has now developed a selective laser ablation technique designed to target PDAC while leaving healthy pancreatic tissue intact.

Thermal ablation techniques, such as radiofrequency, microwave or laser ablation, could provide a treatment option for patients with locally advanced PDAC, but existing methods risk damaging surrounding blood vessels and healthy pancreatic tissues. The new approach, described in Optica, uses the molecular fingerprint of pancreatic tumours to enable selective ablation.

The technique exploits the fact that PDAC tissue contains a large amount of collagen compared with healthy pancreatic tissue. Amide-I collagen fibres exhibit a strong absorption peak at 6.1 µm, thus the researchers surmised that tuning the treatment laser to this resonant wavelength could enable efficient tumour ablation with minimal collateral thermal damage. As such, they designed a femtosecond pulsed laser that can deliver 6.1 µm pulses with a power of more than 1 W.

“We developed a mid-infrared femtosecond laser system for the selective tissue ablation experiment,” says team leader Houkun Liang. “The system is tunable in the wavelength range of 5 to 11 µm, aligning with various molecular fingerprint absorption peaks such as amide proteins, cholesteryl ester, hydroxyapatite and so on.”

Liang and colleagues first examined the ablation efficiency of three different laser wavelengths on two types of pancreatic cancer cells. Compared with non-resonant wavelengths of 1 and 3 µm, the collagen-resonant 6.1 µm laser was far more effective in killing pancreatic cancer cells, reducing cell viability to ranges of 0.27–0.32 and 0.37–0.38, at 0 and 24 h, respectively.

The team observed similar results in experiments on ectopic PDAC tumours cultured on the backs of mice. Irradiation at 6.1 µm led to five to 10 times deeper tumour ablation than seen for the non-resonant wavelengths (despite using a laser power of 5 W for 1 µm ablation and just 500 mW for 6.1 and 3 µm), indicating that 6.1 µm is the optimal wavelength for PDAC ablation surgery.

To validate the feasibility and safety of 6.1 µm laser irradiation, the team used the technique to treat PDAC tumours on live mice. Nine days after ablation, the tumour growth rate in treated mice was significantly suppressed, with an average tumour volume of 35.3 mm³. In contrast, tumour volume in a control group of untreated mice reached an average of 292.7 mm³, roughly eight times the size of the ablated tumours. No adverse symptoms were observed following the treatment.

Clinical potential

The researchers also used 6.1 µm laser irradiation to ablate pancreatic tissue samples (including normal tissue and PDAC) from 13 patients undergoing surgical resection. They used a laser power of 1 W and four scanning speeds (0.5, 1, 2 and 3 mm/s) with 10 ablation passes, examining 20 to 40 samples for each parameter.

At the slower scanning speeds, excessive energy accumulation resulted in comparable ablation depths. At speeds of 2 or 3 mm/s, however, the average ablation depths in PDAC samples were 2.30 and 2.57 times greater than in normal pancreatic tissue, respectively, demonstrating the sought-after selective ablation. At 3 mm/s, for example, the ablation depth in tumour was 1659.09±405.97 µm, compared with 702.5±298.32 µm in normal pancreas.

The findings show that by carefully controlling the laser power, scanning speed and number of passes, near-complete ablation of PDACs can be achieved, with minimal damage to surrounding healthy tissues.

To further investigate the clinical potential of this technique, the researchers developed an anti-resonant hollow-core fibre (AR-HCF) that can deliver high-power 6.1 µm laser pulses deep inside the human body. The fibre has a core diameter of approximately 113 µm and low bending losses at radii under 10 cm. The researchers used the AR-HCF to perform 6.1 µm laser ablation of PDAC and normal pancreas samples. The ablation depth in PDAC was greater than in normal pancreas, confirming the selective ablation properties.

“We are working together with a company to make a medical-grade fibre system to deliver the mid-infrared femtosecond laser. It consists of AR-HCF to transmit mid-infrared femtosecond pulses, a puncture needle and a fibre lens to focus the light and prevent liquid tissue getting into the fibre,” explains Liang. “We are also making efforts to integrate an imaging unit into the fibre delivery system, which will enable real-time monitoring and precise surgical guidance.”

Next, the researchers aim to further optimize the laser parameters and delivery systems to improve ablation efficiency and stability. They also plan to explore the applicability of selective laser ablation to other tumour types with distinct molecular signatures, and to conduct larger-scale animal studies to verify long-term safety and therapeutic outcomes.

“Before this technology can be used for clinical applications, highly comprehensive biological safety assessments are necessary,” Liang emphasizes. “Designing well-structured clinical trials to assess efficacy and risks, as well as navigating regulatory and ethical approvals, will be critical steps toward translation. There is a long way to go.”

The post Resonant laser ablation selectively destroys pancreatic tumours appeared first on Physics World.