The Callascope: A portable, user-friendly, speculum-free device for diagnostic imaging of cervical cancer precursor lesions

Abstract

Cervical cancer remains a leading cause of preventable mortality in low- and middle-income countries, where visual inspection with acetic acid (VIA) is often used for screening. VIA’s reliability is limited by the lack of magnification, absence of imaging for review, and high inter-operator variability. We developed the Callascope, a portable colposcope designed to address these limitations by integrating high-resolution imaging and acetic acid application in a single device. The Callascope can be used with or without a speculum and incorporates an asymmetrical tip for cervix positioning, an integrated camera with illumination, and a built-in atomizer for uniform contrast application. Bench testing confirmed image quality and contrast application were comparable to a standard colposcope. In clinical studies, the Callascope enabled consistent cervical visualization, effective acetowhitening, and reliable image interpretation while improving patient comfort. By addressing VIA’s limitations, it has the potential to expand screening access and support global cervical cancer elimination.

Introduction

Cervical cancer, the 4^th most common form of cancer globally, is caused by persistent infection with oncogenic forms of the Human Papillomavirus (HPV)^1,2. Though cervical cancer incidence and mortality have decreased globally since 2008, these improvements have occurred mostly in high-income countries³ with little or no impact to those living in low- and middle-income countries (LMICs). The main reason for the decline of cervical cancers in high-income countries is the availability of effective screening programs, and more recently, the introduction of the HPV vaccine. While these technologies make cervical cancer almost entirely preventable, the quality, availability, and uptake of these life-saving solutions remain extremely limited in areas where cervical cancer deaths are still prevalent^4–6. As a result, patients present with advanced disease in settings with limited treatment options⁷, leading to over 348,000 deaths each year^8,9. The number of expected deaths will continue to increase in LMICs and is expected to double by 2050¹⁰.

In high-income countries, identifying and treating patients with precancer requires multiple steps. Patients first undergo screening with a Pap smear, HPV test, or co-testing with both. If results are abnormal, they return to clinic for colposcopy and biopsy to confirm precancerous changes^11,12. During a colposcopy exam, a speculum is inserted to open the vaginal canal, and contrast agents such as acetic acid and, depending on clinical practice, Lugol’s iodine are applied to the cervix. Acetic acid coagulates nuclear proteins; because dysplastic cells have higher nuclear-to-cytoplasm ratios than normal cells, this effect is more pronounced in abnormal tissue, producing the characteristic acetowhite appearance¹³. Lugol’s iodine stains glycogen-rich normal squamous epithelium dark brown, while pre-cancerous cells, which lack glycogen, remain unstained and appear pale yellow¹⁴. These contrast-enhanced changes are then visualized with a colposcope using varying magnification and optical filters, and biopsies are obtained for histopathological review. If precancer is confirmed, patients then undergo a third visit for treatment, such as ablation or a Loop Electrosurgical Excision Procedure (LEEP), as necessary. This model has significantly reduced cervical cancer incidence in high-income countries but relies on costly equipment, infrastructure, and multiple visits, making it impractical in LMICs^12,15.

A simplified process called visual inspection with acetic acid (VIA) is often used as the primary screening tool in LMICs¹⁶. During VIA, assessment of precancer risk is done by observing the cervix through the speculum without imaging or biopsy confirmation, relying on the unaided eye to detect acetowhite changes. Although inexpensive, VIA suffers from significant limitations: reliance on the speculum, lack of magnification, inconsistent lighting, and high inter-operator variability^5,17–19. These factors contribute to wide variation in the sensitivity and specificity of VIA across different providers and clinical contexts. Colposcopy offers magnification and biopsy confirmation but requires costly equipment, multiple visits, and infrastructure that is not widely available. In addition to these technical constraints, many patients avoid screening because of discomfort with the pelvic exam itself. Cultural norms regarding modesty, prior experiences of sexual trauma, and pain associated with speculum insertion and manipulation can discourage patients from attending cervical cancer screening visits^18,20,21.

The introduction of portable imaging devices, which would serve as alternatives to traditional colposcopes, has the potential to address image related factors such as low-lighting and lack of magnification in LMICs^22–25. However, even with the availability of imaging, multiple steps in the exam may affect image quality, particularly when performed by less experienced providers or in settings with high provider turnover²⁶. Obtaining an unobstructed view of the entire cervix requires careful positioning of the speculum and camera, steps that are influenced by both the device design and operator technique. Additionally, contrast enhancement achieved with acetic acid is time-dependent, peaking between 1–2 min²⁷. Yet the timing of assessment, either by direct observation or imaging, differs between providers and clinical sites, contributing to variability in the identification of suspicious areas.

To address these challenges, we describe in this manuscript the development and clinical testing of the Callascope, designed to consolidate cervical imaging and contrast application into a single, portable device that can be used either with or without a speculum. Building on our prior work with the Pocket Colposcope and the original Callascope^{18,22,23,28–32}, this redesigned device incorporates a calla lily-shaped tip and thin shaft for comfortable insertion, an integrated 5-megapixel camera with white LEDs for high-quality imaging, and a built-in atomizer to deliver acetic acid without device removal. In speculum-free operation, the vaginal wall stabilizes the device to minimize motion artifacts, while autofocus accommodates small changes in distance from the cervix. This technology has the potential to impact cervical cancer detection by addressing key limitations of VIA; the Callascope aims to improve patient comfort by eliminating the need for the speculum while maintaining the image quality and uniformity necessary for reliable diagnostic interpretation.

Results

Callascope design

The Callascope is composed of a slim, tubular body that houses the camera and atomizer along with a flared tip to help the user center the cervix, allowing optimal visualization without the need for a speculum (Fig. 1A). To accommodate patients with varying Body Mass Indices (BMI) and parities, the device includes removable tips that can be swapped out according to individual comfort and size. The body of the Callascope is 3D printed in a sterilizable medical grade resin (FormLabs BioMed Black Resin³³). This design sought to reduce the discomfort commonly associated with traditional gynecological tools and provides a more participant-friendly experience. All materials and components of the device are easy to source and can be manufactured in different quantities.

Fig. 1. Design of the Callascope and atomizer.

A A photo of the Callascope device with a slender and tampon-like design, featuring a removeable Calla lily-shaped tip enclosing the camera. B A 5MP camera and LED ring protected by a hydrophobic window with the addition of an atomizer device to spray contrast onto the cervix. C A photo of the 5MP camera and LED ring integrated into the body of the device, covered by the optically transparent hydrophobic window. D The axial whirl atomizer with both CAD and a photo of the 3D printed component. Contrast is pressurized through the axial whirls and forced out a small exit orifice to create a conical spray distribution on the cervix.

The key optical components of the Callascope are shown in Fig. 1B. The device features a 5MP camera (Part No. CMT-5MP-OV5693-S571, Shenzhen ChuangMu Technology Co) with an integrated ring of four white LEDs at its tip, providing uniform illumination of the cervix and vaginal walls (Fig. 1C). These LEDs are protected by an optically transparent hydrophobic window (Part No. 34–524, Edmund Optics), which prevents moisture buildup and fogging, ensuring clear visualization in the humid environment of the vaginal canal. The hydrophobic window is bonded to the Callascope body with a medical grade epoxy resin that enables the device to be waterproof. The camera module is wired to a micro processing board that controls the LED brightness and connects the camera to a USB board. A female-to-male USB adapter is used to connect the USB board to a phone or tablet.

Figure 1B also shows an axial whirl pressure atomizer inserted through the tip of the Callascope. The atomizer has a maximum diameter of 4.75 mm and is 11 mm in length to fit within the size constraints of the Callascope. Atomizers are 3D printed in a biocompatible and sterilizable resin (FormLabs BioMed Amber Resin³⁴). It is connected to a tubing system, 3 mm in diameter running along the side of the device to deliver a contrast agent directly to the cervix. The axial whirl atomizer consists of two helical channels, a mixing chamber, and an exit orifice to provide uniform coverage without pooling of the liquid (Fig. 1D). By producing a tangential component to the fluid velocity along with a pressure drop along the atomizer orifice, the atomizer ejects the fluid into a cone of droplets that evenly cover the tissue surface of the cervix. Tilting the tip of the Callascope at a 32° angle relative to the optical axis resulted in maximal spray coverage of the cervix without infringing on the camera FOV. Both the atomizer and the plastic tubing are disposable and replaced between patients, while the external shell is reusable and cleaned by a standard high level disinfection protocol. The same tubing system can be used sequentially for acetic acid and Lugol’s iodine within a single exam. The redesigned Callascope was tested to verify adequate image quality, waterproofing, and spray coverage of the cervix by the atomizer.

The Callascope presented here represents an advancement of the two prior devices developed by our group. The original Pocket Colposcope was a speculum-based device that required insertion through a standard speculum for cervical visualization, relying on a 5MP camera and external lighting at a short working distance^22,23,31,32. The original Callascope introduced a speculum-free, patient-centered design with a calla lily-shaped tip to center the cervix without a speculum, but had a 2MP camera and no integrated optimized atomizer^18,28,29. The current Callascope builds on this foundation with three key advances: (1) a new 5MP camera module providing the same image resolution to the Pocket colposcope; (2) an integrated axial whirl atomizer for in-situ uniform contrast agent delivery directly to the cervix, eliminating the need for a separate applicator; and (3) a removable, interchangeable tip/sizing system to accommodate varying patient anatomy and BMI. A comparison of imaging specifications across the standard colposcope, the Pocket Colposcope and the current Callascope is provided in Table 1.

Table 1.

Comparison of imaging specifications for the Callascope, the Pocket Colposcope, and a standard-of-care colposcope (Leisegang Optik 2)

Device Characteristic		Standard-of-Care Colposcope	Pocket Colposcope	5MP Callascope
White/Green Light		White + Green	White + Green	White
Resolution (MP)		18	5	5
Resolving Power (μm)		21–51	15–30	25–40
Diagonal Field of View (mm)		19–77	35–45	38–52
Depth of Field (mm)		7.1–19.6	8.5–10.9	6–20
LED Power (lux)		22,173	600–5000	1200–7100
Mean Color Accuracy:	ΔC*ab	2.9	9.9	7.4
	ΔE*ab	5.2	12.2	10.5
Weber’s Contrast Ratio		2.1	1.7	2.4
Contrast Coverage		100%	100%	95.4%

Camera verification testing

The Callascope demonstrated comparable resolution, diagonal field of view, depth of field, contrast, and distortion to the Leisegang Optik 2 colposcope using previously developed methods for verification testing^22,23. Detailed results with representative test images are provided in the supplemental materials (Fig. S1), and a summary of image quality metrics is presented in Table 1.

Atomizer testing and integration

The atomizer in the Callascope uses less than 0.4 mL of fluid to prevent pooling in the vaginal canal and functions under operating pressures generated by 3 mL and 6 mL syringe sizes (minimum of 70 psi). The atomizer (Fig. 1D) was shown to reliably coat at least 95% of the cervix area with Lugol’s Iodine liquid contrast in bench-testing on a mock-cervix with a 33 mm diameter (Fig. S1E). The average coverage of the cervix was 98.6%, with a standard deviation of 0.91% total coverage, when contrast was applied to the mock-cervix without the tip present and with the atomizer body held in place by a 3D printed fixture. When integrated into the tip at a 32° angle relative to the optical axis, the coverage did not change significantly, with an average coverage of the mock-cervix of 95.41% and a standard deviation of 8.9% coverage. No significant differences were found between the coverage with and without the inserter tip (two-tailed Wilcoxon rank-sum, independent, with an alpha level of 0.05).

Waterproofing Verification

The maximum exam length is expected to be 10 min, with a device reprocessing time, including submersion in disinfectant, for approximately 30 min. The device was validated to be waterproof at 30 min, one hour, and eight-hour submersions. The Callascope was not found to have any visible water ingress after these submersions, and the camera and LEDs remained functional with unchanged image quality after repeated submersion testing. The atomizer is a disposable component and was not included in this testing.

Clinical study results

A two-arm clinical study was conducted at Duke University Medical Center to assess the acceptability and function of the Callascope. The first arm assessed comfort, image quality, ease-of-use, and functionality of a speculum-free exam in participants scheduled to undergo a Pap smear collection (imaging-only participants). The second arm assessed imaging and contrast agent delivery with the atomizer in participants scheduled to undergo colposcopy (contrast study participants). In this arm, one group of participants underwent contrast application and imaging with the Callascope through a speculum, and a separate group underwent speculum-free insertion. A summary of the workflow and data collected in each study arm is shown in Fig. 2 for reference.

Fig. 2. Clinical study description and data collection summary.

Two study arms evaluated 1) the comfort of the device for imaging-only participants and the ability to capture speculum free images, and 2) the ability to apply contrast to the cervix with the atomizer both with and without the speculum.

Clinical study demographics

Participant demographics for the two arms of the study are summarized in Table 2. Included are responses from participants in the speculum-free imaging arm, who completed a questionnaire before the exam capturing their history with speculum exams and whether they viewed the speculum as a barrier to care. Physicians completed post-exam surveys for 19 exams.

Table 2.

Demographic information for study participants in both arms: speculum-free imaging-only (n = 20) and contrast application (n = 27)

		Imaging-Only Participants	Contrast Study Participants
Age (years)	Mean (SD)	35.5 (7.3)	39.4 (10.5)
	Min	26	22
	Max	55	63
Race	Asian	0 (0%)	1 (3.7%)
	American Indian	0 (0%)	1 (3.7%)
	Black or African American	7 (35%)	9 (33.3%)
	Other	2 (10%)	0 (0%)
	White	11 (55%)	16 (59.3%)
Ethnicity	Hispanic	2 (10%)	2 (7.4%)
	Not Hispanic	17 (85%)	23 (85.2%)
	Unknown	1 (5%)	2 (7.4%)
BMI	18.5–24.9	9 (45%)	7 (25.9%)
	25–29.5	4 (20%)	4 (14.8%)
	>30	7 (35%)	16 (59.3%)
Number of vaginal births	0	13 (68.4%)
	1	1 (5.3%)
	2	3 (15.8%)
	3	1 (5.3%)
	4	1 (5.3%)
Regular user of tampons/ menstrual cups	Yes	12 (63.2%)
	No	7 (36.8%)
Prior number of speculum examinations	1–2	0 (0%)
	3–5	6 (33.3%)
	6–10	2 (11.1%)
	>10	10 (55.6%)
	Not answered	1 (5.6%)
Perception of speculum as a barrier	No - small barrier	11 (57.9%)
	Medium - large barrier	5 (26.3%)
	Not answered	3 (15.8%)

Prior experience and variables related to experience with pelvic exams in imaging-only participants are also reported.

Speculum-free imaging

In the first arm of the study, the Callascope was used to validate the comfort and image quality of the redesigned device in a speculum-free exam of 20 participants scheduled for Pap exams. We have shown that a speculum-free exam with previous iterations of the Callascope significantly improves patient comfort compared to a speculum-based exam^18,28–30, so the results of this study arm are abbreviated.

A total of 20 participants were imaged after pap smear collection (3 participants imaged both with and without the speculum at an appropriate focal length, and 17 additional participants with images only from the Callascope without the speculum). Figure 3A shows the three pairs of Callascope images obtained with and without the speculum that had comparable fields of view. In both imaging modalities for the leftmost example, the cervix is captured at slightly oblique orientations, highlighting how natural variations in cervical positioning can occur independent of the imaging format. Blood is visible in both speculum-free and through-speculum images for the middle and right example, indicating that these obstructions arose from the patients’ menstrual cycle rather than from the choice of imaging approach. Despite these features, the Callascope was able to capture comparable views of the cervix across both modalities, demonstrating consistent image quality and visualization of clinically relevant landmarks.

Fig. 3. Cervical image quality and patient experience during speculum-free Callascope exams.

A Representative image pairs captured from speculum-free imaging-only participants during physician insertion of the Callascope at similar working distances without a speculum (left) and through the speculum (right). These paired images show the comparable fields of view and cervix orientation with and without the speculum in place. B Post-exam questionnaire examining overall participant preference for the Callascope exam compared to the speculum exam. 100% of participants preferred the Callascope to the speculum.

Of the twenty patients, 18 had images that were in focus. All four quadrants of the cervix were visible in nine patients. In the other 9 focused images, the os was not centered so that less than four quadrants were visible. The remaining two images were blurred; one of these had four quadrants visible, and one was tilted. Biological features partially obstructed eight images, with menstrual blood covering the cervix in two participants, and mucus covering some features of the cervix in six additional participants. Images with less cervix visibility were often (5/9 participants) due to bleeding caused by the Pap smear collection, or challenging anatomy like a flush or tilted cervix.

Post-exam surveys indicated minimal discomfort, with 90% of participants rating every phase of the exam at 3 or lower on the Wong-Baker pain scale (0–10). Mean pain scores were 1.2 for insertion, 0.65 for manipulation, and 0.40 for removal, respectively, Fig. S2). Participants were also asked about their overall experience with the Callascope compared to the speculum. After having both devices used on them, participants overwhelmingly favored the Callascope, with 50% of patients reporting the Callascope exam was “much better” and 50% finding it “slightly better” than the traditional speculum exam (Fig. 3B). No patients reported that the Callascope experience was worse than the speculum. Additionally, physicians expressed high confidence and ease-of-use with the device, even with minimal prior experience (Table S1).

Contrast-agent application with the atomizer

Contrast agent application and image quality

Images or videos were captured pre- and post-acetic acid application with the Callascope in 26/27 participants immediately prior to standard colposcopy. Acetic acid was applied with the atomizer through the speculum in 26 participants (23 with images, 3 with videos), and speculum-free in 4 participants. The Callascope was unable to image the cervix in one patient as it did not fit through the speculum. Refer to Table 2 for demographic data on these participants. Of the 22 participants with images or videos captured by the Callascope through the speculum, three did not have corresponding post-acetic acid standard of care (SOC) images due to a malfunction with the image-saving feature of SOC colposcope (19 Callascope images with 16 corresponding SOC colposcope images, 3 Callascope videos with 3 corresponding SOC images). Additionally, there was a national Lugol’s iodine shortage³⁵ for part of the study. Therefore, post-Lugol’s iodine images were obtained from eight participants with the Callascope used through the speculum. One additional image post-Lugol’s iodine was captured in the speculum-free format. SOC colposcope images were obtained from only seven of these patients due to the malfunction previously mentioned (6 paired to through-the-speculum Callascope images, 1 paired to a speculum-free video). The os was visible in all images.

The proportion of exams achieving full four-quadrant visualization was comparable between the Callascope and SOC colposcope imaged through the speculum (Callascope: 70.4%, SOC: 82.6%). In 16 of 21 paired observations, both devices achieved the same quadrant count. Across all 21 paired images, mean quadrant scores did not differ significantly between the Callascope and standard colposcope (Wilcoxon signed-rank, p = 0.655) for either acetic acid or Lugol’s iodine imaging (3.71 vs 3.76). These findings suggest comparable cervical visualization between devices when the Callascope was used through the speculum (Table 3).

Table 3.

Cervical quadrant visibility for the Callascope and SOC colposcope in the through-the-speculum contrast arm

		4 Quadrants	3 Quadrants	<3 Quadrants
Callascope	Post-Acetic Acid (n = 19)	14 (73.7%)	4 (21.1%)	1 (5.3%)
	Post-Lugol’s Iodine (n = 8)	5 (62.5%)	3 (37.5%)	0 (0%)
	Total (n = 27)	19 (70.4%)	7 (13.0%)	1 (3.7%)
Colposcope	Post-Acetic Acid (n = 17)	13 (76.5%)	3 (17.6%)	1 (5.9%)
	Post-Lugol’s Iodine (n = 6)	6 (100%)	0 (0%)	0 (0%)
	Total (n = 23)	19 (82.6%)	3 (13.1%)	1 (4.3%)

Full four-quadrant visualization was achieved in 70.4% of Callascope exams and 82.6% of SOC colposcope exams; ≥3 quadrants were achieved in 96.3% and 95.7% respectively. No statistically significant difference was found between devices across 21 paired observations (mean quadrant scores 3.71 vs 3.76; Wilcoxon signed-rank, p = 0.655).

Contrast was observed to effectively cover the cervix in 26/27 contrast study participants. In the first participant, acetic acid was re-applied because only 0.4 mL of acetic acid was sprayed with the atomizer due to bench-testing results before it was determined that coverage was insufficient. 1 mL volumes were used for all subsequent applications. Effective coverage was determined by the physician based on the exam context. For example, in 14/26 cases with effective coverage, physicians were able to biopsy suspicious locations from atomizer contrast application alone.

In another 9/26 cases, more acetic acid was applied only because it was suspected that mucus or blood may have impeded uptake in a specific region, or that the acetowhitening effect may have faded before the physician was ready to biopsy. 3/9 cases in this group were from speculum-free insertion of the Callascope. In all speculum-free videos, contrast coverage was deemed sufficient and can be seen covering the full cervix in video capture, though mucus occluded the edges of the image in one of these three patients. More acetic acid was applied after the insertion of the speculum due to fading of the acetowhitening effect and to ensure that the mucus obstructions had not inhibited coverage. The transformation zone was also not fully visible in these cases, so acetic acid was applied through the speculum to cover the slightly wider field of view.

The final 3/26 participants had acetic acid re-applied because they had cervixes that appeared near-normal after acetic acid was initially applied through the atomizer. The acetic acid was re-applied to verify that the normal appearance was not due to a device malfunction or insufficient coverage. Slight acetowhitening that led to a biopsy was only seen in one of these three cases, which was also observed in the same region through the Callascope.

Representative images demonstrating contrast-induced changes to the cervix are shown in Fig. 4. Imaging was performed using both the Callascope and SOC colposcope following contrast delivery with the atomizer through the speculum with representative images from patients where a biopsy was taken, but histology was negative (Fig. 4A). Areas of acetowhitening and Lugol’s iodine-negative regions corresponded with each other and with those observed in the SOC images. Biopsy locations (marked with arrows) aligned with post-hoc physician-annotated regions of concern (marked with circles). Figure 4B shows full Lugol’s iodine coverage in both use cases of the Callascope– speculum-based (top) and speculum-free (bottom). In the example through the speculum (top), no biopsy was taken, consistent with uniform dark Lugol’s staining and a visually normal cervix. Iodine-negative areas (bottom) do not necessarily indicate pre-cancerous change; columnar epithelium, which lacks glycogen, will appear unstained on Lugol’s application in the absence of any disease¹⁴.

Fig. 4. Representative images captured pre-, post-, and throughout contrast application with the atomizer.

A All images in Panel A are from the same patient. Negative histology from biopsies taken at eight and one O’clock, annotated post- acetic acid and Lugol’s iodine application with the Callascope through the speculum. An IUD string is visible throughout the image series and is indicated by a white arrow on the pre-acetic acid Callascope image. Acetowhite regions correspond to the areas of the Lugol’s images that did not stain fully brown. There was comparable image quality between images taken with the Callascope (top) and images taken with the SOC colposcope (bottom). Biopsy locations are indicated with arrows, while circles indicate a single annotation per image from the attending physician. B Pre- and post- Lugol’s iodine images from two different patients captured with the Callascope: the top row shows a through-the-speculum exam in a visually normal cervix where no biopsy was taken, and the bottom row shows a speculum-free exam from a separate patient with negative histology. In the bottom row, the inferior portion of the cervix does not take up Lugol’s stain, consistent with columnar epithelium, which lacks glycogen and does not stain regardless of disease status, as confirmed by the attending physician. Lugol’s staining otherwise shows full contrast coverage over the cervix.

To compare Callascope image quality with that of the SOC colposcope, post-acetic acid images were evaluated using a blur-detection algorithm developed in our lab as an objective measure of image sharpness/ blur³⁶. Lugol’s iodine images were manually reviewed for blur, since the algorithm was not trained on this contrast type (Table 4). Videos frames were not analyzed. Since no images were captured pre-acetic acid application with the SOC colposcope in accordance with standard procedure, only post-contrast images are used to compare the device image quality. A larger proportion of colposcope images were blurry compared to that with the Callascope; 9/23 images were blurry with the Colposcope compared to 5/27 blurry with the Callascope, though this difference was not statistically significant (χ² = 1.695, p = 0.193) (Table 4). No pooling of contrast occurred that obstructed images.

Table 4.

Summary of blurry vs clear images captured with the Callascope and Colposcope after contrast application

		Clear	Blurry
Callascope	Post-Acetic Acid (n = 19)	17 (89.5%)	2 (10.5%)
	Post-Lugol’s Iodine (n = 8)	5 (62.5%)	3 (37.5%)
	Total (n = 27)	22 (81.5%)	5 (18.5%)
Colposcope	Post-Acetic Acid (n = 17)	11 (64.7%)	6 (35.3%)
	Post-Lugol’s Iodine (n = 6)	3 (50%)	3 (50%)
	Total (n = 23)	14 (60.9%)	9 (39.1%)

Percentages shown were calculated based on the total images for the specific contrast agent listed. 18.5% of the total post-contrast images captured with the Callascope were blurry, compared to 39.1% of post-contrast Colposcope images (not significant, p > 0.05; Cohen’s κ = −0.197).

When assessed at the patient level, agreement between devices on whether images were classified as blurry or clear was poor (Cohen’s κ = −0.197), suggesting that the devices did not consistently capture images of comparable clarity within the same patient. Taken together, while both devices produced interpretable images, the Callascope demonstrated fewer blurry images overall, although this difference did not reach statistical significance (Table 4).

Clinician interpretation and concordance with biopsy sites

Two fourth year residents and one attending gynecologist independently completed post-hoc surveys and annotations of the first 17 sets of participant images captured with the Callascope. Table 5 describes the results of the physician annotations. Of the 17 participants with annotated images, 1–3 biopsies were obtained in 13 of them at the time of standard colposcopy. At least one lesion was identified as a potential biopsy site in all participant images by all physicians. On average, physicians were able to use Callascope images to correctly annotate at least one biopsy location in 84.6% of the 13 participants from whom biopsies were obtained. A total of 4/13 participants who were biopsied had histology-confirmed disease. Across all cases with abnormal histology, the attending physician and one resident successfully identified at least one region for biopsy based on Callascope images. The other resident only missed the biopsied lesion location in 1 of the 4 participants with abnormal histology.

Table 5.

Agreement between clinician annotation and actual biopsy locations

Patients with biopsies taken		13
Number of normal histology patients		9
Normal histology patients where at least one biopsy was correctly annotated	R1	7 (77.8%)
	R2	7 (77.8%)
	A1	8 (88.9%)
	Average	7.33 (81.5%)
Number of abnormal histology patients		4
Abnormal histology patients where at least one biopsy was correctly annotated	R1	4 (100%)
	R2	3 (75%)
	A1	4 (100%)
	Average	3.67 (91.7%)

R1 and R2 were fourth-year residents, A1 was an attending physician in obstetrics and gynecology. Annotations were compared to collected biopsy sites for each patient. Overall, for 84.6% of participants, clinician annotations matched at least one biopsy site. Agreement was 81.5% for participants with normal histology and 91.7% for participants with abnormal histology, indicating comparable performance across histology outcomes.

In the 9/13 participants with normal histology, the two residents and the attending had an 81.5% success rate in identifying at least one location that biopsy was taken in clinic (Table 5). If the results from all three readers were combined (or a combination of the attending with either resident), at least one lesion that was biopsied in clinic would have been identified in each participant. Results from the attending physician on a per-biopsy basis are available in the supplemental materials (Table S2).

Temporal and quantitative assessment of acetowhitening

While clinician annotations of Callascope images showed high agreement with biopsy locations, visual interpretation alone may be influenced by reader experience, lesion subtlety, or image variability. To complement these qualitative assessments, the temporal behavior of acetowhitening following acetic acid application was assessed in the images and video frames for cases in which biopsies were taken, quantifying regions of interest in the pre- and post-contrast images (n = 18).

Time series images from Callascope recordings illustrate the temporal dynamics of acetowhitening following acetic acid application, captured both through the speculum and in the speculum-free format (Fig. 5A, B). Regions of interest were selected at biopsy sites, adjacent visually normal regions, and in some cases ectropion. While representative video still frames with quantified regions of interest are shown for individual cases to depict the whitening response over time (Fig. 5C, D), only the pre-acetic acid image and the final post-acetic acid image were used for the full quantitative analysis across all cases. In these examples, biopsy regions demonstrated either little change or sustained whitening, while normal regions in both examples remained relatively unchanged (Fig. 5C, D); interestingly, a selected ectropion region whitened rapidly and then faded, while whitening at the lesion site persisted (Fig. 5C). Across all cases, the mean RGB intensity in the post-acetic acid image alone was calculated for each region of interest (Fig. 5E) and compared to the mean percent change in normalized RGB intensity between the two time points (Fig. 5F). Static post-acetic acid intensity values alone did not differ significantly between tissue regions (Fig. 5E). However, when the percent change in intensity between the pre- and post-contrast images was considered, biopsy regions had a significantly higher intensity than normal-appearing regions (p < 0.001, paired t-test; Fig. 5F). In 16/18 cases, regions where biopsies were taken exhibited a positive percent change in intensity that was greater than adjacent visually normal regions.

Fig. 5. Temporal imaging and quantitative analysis of acetowhitening responses with the Callascope.

Representative time series images captured with the Callascope both A through the speculum and B in the speculum-free format. Blue boxes indicate biopsy sites, orange boxes indicate adjacent visually normal regions, and yellow boxes indicate ectropion tissue. C Cropped regions of interest from the series captured through the speculum and corresponding mean percent changes in the normalized RGB intensity highlight whitening responses over time: biopsy and normal regions remain relatively unchanged, and ectropion shows rapid initial whitening that fades quickly. D Cropped regions of interest from the speculum-free example showing rapid and sustained whitening in the biopsy region while the normal tissue fades quickly back to baseline intensity. E Mean RGB intensity measured from the static post-acetic acid image alone showed no significant difference between biopsy and normal regions. F When the change was normalized to the pre-contrast image, regions where a biopsy was taken exhibited significantly greater percent change in intensity compared to visually normal regions (paired t-test, ***p < 0.001).

Discussion

This study demonstrates the performance, acceptability, and usability of the Callascope with participants in either the speculum-free imaging arm (Arm One) or the contrast application arm (Arm Two). We demonstrated that the Callascope produces image quality comparable to that of a standard clinical colposcope, which is orders of magnitude larger, heavier, and more costly. In addition, the Callascope provides unique capabilities including a simplified exam procedure that minimizes the potential for inter-operator variability, improved comfort compared to a speculum-based exam and a speculum-free option, which provides patients with the flexibility to perform their own exam outside of a standard clinic. The improved comfort reported by patients is consistent with our prior publications even when early prototypes of the Callascope were used^18,28–30.

Given the Callascope’s potential for use in low-resource settings where VIA remains the dominant screening modality, it is worth noting that VIA has significant provider variability; because colposcopy is the imaging benchmark for cervical cancer diagnosis, it serves as the appropriate comparator for initial validation of the Callascope. Prior work from our group has established that the Pocket Colposcope performs comparably to standard-of-care colposcopy for detecting cervical precancerous lesions across multiple international sites^31,32. Triaging VIA-positive women with the Pocket Colposcope also significantly reduced overtreatment and improved optimal treatment rates compared to VIA alone³⁷. While a direct comparison to VIA was not performed in this study, evaluating Callascope performance within VIA-based screening workflows remains an important next step for translation to low-resource settings.

The Callascope captured high-quality, speculum-free images in the speculum-free imaging-only participants. Only two of twenty participants had blurred images, primarily due to an early tip iteration printed in white that interfered with autofocus. This issue was resolved by switching to a black tip, after which no further blur occurred. Providers using the Callascope were able to visualize the cervical os in all participants. More than half of participants did not view the speculum as a significant barrier to screening prior to the exam and rated the Callascope similarly to the speculum in willingness to use it (Fig. S3). Despite this initial hesitancy, 100% of participants preferred the Callascope to the speculum after the exam, likely due to comfort across all phases of the exam. It should be noted however that the speculum-based exam prior to the insertion of the Callascope may have also influenced participants’ perceptions; participants may have felt relatively less discomfort during the Callascope exam. The willingness and preference of patients in this study to use the Callascope in different situations is supported by prior self-exam studies conducted by our group in Ghana, the U.S. and Peru^18,28–30. Physicians also reported that the Callascope was straightforward to use and manipulate, with high overall confidence even among providers with limited prior experience. However, in some cases, the providers did not continue to manipulate the Callascope to center the os if the patient was bleeding from their Pap exam or had difficult anatomy like a flat or tilted cervix. More training may address this and build physician confidence with the asymmetrical tip.

In contrast study participants, both image quality and acetic acid coverage were compared to standard of care. Our findings demonstrate that atomizer-based acetic acid application can produce high-quality images and acetowhitening in areas of histologically confirmed disease. The absence of pooling from contrast application underscores the atomizer’s effectiveness in evenly applying contrast to the cervix without obstructing the camera’s view, and physician confidence in acetic acid coverage was high throughout the study. Notably, we demonstrated the feasibility of speculum-free contrast application and imaging simultaneously; the Callascope is unique in its ability to integrate both functions in a single device. Image quality was strong overall, with fewer blurry images observed with the Callascope than with the SOC colposcope (18.5% vs 39.1%), though this difference was not statistically significant (p > 0.05, Cohen’s κ = −0.197) and likely reflects study-specific factors rather than typical colposcope performance. Despite this, several image quality challenges were encountered, though none broadly impacted overall image interpretation. Lugol’s iodine application occasionally introduced haze, likely due to the darker contrast interfering with the camera’s auto-gain, which can be addressed with anti-fog wipes or gain adjustment. Cervical mucus and blood occasionally reduced image quality, though these challenges are not unique to the Callascope and are commonly encountered during standard speculum exams. However, because the Callascope does not currently include tools for clearing mucus, this issue may present a barrier to its usability in some cases. When visualization was significantly compromised in the speculum-free arm, providers switched to a through-the-speculum approach as an alternative. To address this limitation, we have begun development of a simple finger-cot–like tool for biological obstruction removal that will allow patients to gently swab their cervix prior to inserting the Callascope, and ongoing studies are evaluating cotton swab pre-clearance as an interim solution. Patients who are actively menstruating should be advised to reschedule, as heavy bleeding represents a relative contraindication to use.

Evidence from several studies supports that intermittent imaging of the acetic acid kinetics on the cervix over a 1–2-min window enhances lesion detection^38–40, highlighting the importance of capturing images across this timeframe. The Callascope can capture co-registered pre- and post-contrast images without device removal, enabling time-series evaluation of acetowhitening dynamics. Initial cases show that regions where biopsies were taken demonstrated sustained whitening, while adjacent visually normal regions showed little or no change, and ectropion areas whitened rapidly but faded quickly. Quantitative analysis confirmed that biopsy-positive regions exhibited a significantly greater percent change in intensity than normal tissue, underscoring the value of temporal imaging in highlighting clinically relevant differences. Furthermore, the ability to apply acetic acid or Lugol’s iodine without interrupting imaging can allow co-registration of acetic acid and Lugol’s iodine images, which has been shown to improve the diagnostic accuracy of machine learning algorithms²⁴. Future work will assess videos captured by the Callascope and how changing color and texture features over the course of the acetowhitening effect can be used to perform unsupervised segmentation of pre-cancerous lesions or abnormalities. This will support real-time decision-making and reduce operator variability, particularly in low-resource settings.

The post-hoc annotation study suggests that both residents and the attending physician were generally able to identify clinically relevant areas of concern, with strong alignment between annotated regions and biopsy-confirmed pathology. At least one biopsy site was correctly identified in nearly all cases, which supports the Callascope’s intended role as a screening tool rather than a definitive diagnostic instrument. Some variability between annotations and biopsy locations was expected given the subjective nature of colposcopic exams, and results would be expected to improve with further training⁴¹. Additionally, differences in the image fields of view and annotation methods may have contributed to this variability. First, the images were all oriented by hand to ensure similar orientation of the Callascope and colposcope images, which may have had a slightly different field of view; second, biopsy locations recorded from clinic relied on clock face positions which can be subjective and imprecise. Though biopsies cannot be obtained without removing the Callascope, this is not applicable in many global health contexts where cervical biopsies are not routinely performed during screening. Again, the availability of the temporal data could be leveraged to aid with annotation and biopsy or as noted earlier with automated diagnosis in instances where biopsies are not taken. This feature also provides a powerful training tool, allowing new providers to visualize cervical changes before and after contrast in a single, continuous field of view, thereby accelerating training and aiding in image interpretation.

From a global health perspective, the Callascope could represent a critical step toward achieving equity in cervical cancer prevention. The World Health Organization has emphasized the importance of innovative approaches to improve screening coverage and reduce disparities in cervical cancer outcomes¹⁰. The simplicity of the Callascope will allow community health workers to either perform the procedure or guide a self-exam in a non-clinical setting. This will allow more than 95% of the population to complete screening without the need for a professionally licensed health provider, health facility, or extensive training, dramatically reducing the number of patients who would be expected to seek care at a health facility^15,42. This will allow vital resources like transportation, scheduling appointments, provider time, and materials to be directed to those who need immediate care. Coupled with machine learning algorithms, the decision-making process can be made reliable, consistent, and efficient when expert readers are unavailable^24,43.

Methods

This study consisted of multiple components, including benchtop device validation and a three-arm clinical investigation evaluating the Callascope’s user acceptability, imaging performance, and contrast delivery.

Image and device quality testing

The resolution, Weber’s contrast, color accuracy, LED brightness, diagonal field of view (DFOV), and depth of field (DoF) of the Callascope camera were systematically tested and compared to the $15,000 Leisegang Optik 2 standard-of-care colposcope, following methods described by Lam et al., 2015 and 2018^22,23. Each of the verification tests described were performed multiple times (n = 6).

Resolution was evaluated by imaging a USAF 1951 resolution target under controlled conditions. The smallest resolvable parallel line pairs that still had a 2x difference in pixel intensity were identified using FIJI⁴⁴ for each image. The resolution of these features in microns was determined using the USAF 1951 interpretation table. For Weber’s contrast, pixel intensity values from images of mock cervical lesions (white, $I_{\max }$) and non-lesion background regions (reddish pink, $I_{\min }$) were measured using Fiji open-source software. The contrast was calculated using the formula $\frac{I_{\max }-I_{\min }}{I_{\min }}$.

Diagonal field of view (DFOV) was estimated by placing the cameras at a fixed working distance (WD) from the USAF 1951 resolution target. Using the captured images, the horizontal and vertical dimensions of the visible target area were measured in pixels. These pixel measurements were converted into physical dimensions (in millimeters) using the target’s known element sizes. The Pythagorean theorem is then applied to measure the DFOV of the images at a given WD. Depth of field (DoF) was measured by imaging a 5–15 DoF target (Edmund Optics⁴⁵). FIJI was used to perform a line scan on the sharpest edge of the target in each image, plotting intensity profiles to identify peaks and troughs. The depth of field was calculated as the distance between the nearest and farthest points where the image’s successive peaks and troughs maintained a 2x difference in intensity.

Additionally, color accuracy was measured to evaluate the Callascope’s ability to reproduce true colors by imaging a calibrated color target with known reference values (X-Rite Rez Checker, Edmund Optics⁴⁶). The captured images were analyzed using the Imatest software⁴⁷ to compare the recorded colors against the known reference values of the target. The color reproduction error with luminance difference ($\mathrm{\Delta }{E}_{\mathit{ab}}^{*}$) and color reproduction error ($\mathrm{\Delta }{C}_{\mathit{ab}}^{*}$) were both quantified. LED brightness at both high and low settings were assessed by measuring light intensity at a fixed distance using a photometer (BK Precision 615 Light Meter⁴⁸) and light integration sphere, reporting results in lux.

To verify that the Callascope is waterproof, both to withstand examination fluids and high-level disinfection between exams, it was submersed in water for 30 min (three times the estimated longest length of an exam based on prior studies³⁰). After removal, the Callascope main body was inspected for any signs of water ingress, and a resolution target image was analyzed to ensure the camera and LED function.

Atomizer spray testing

Atomizer spray coverage was evaluated with a custom bench-testing set-up utilizing cotton pieces in a 3D cervix model with a screw-on vaginal wall. The 3D mock-cervix creates a 33 mm diameter 2D image of the cervix when projected into the camera’s field of view (FOV). Different syringe sizes, fluid volumes, and atomizer designs were tested to optimize the coverage of Lugol’s iodine onto the mock-cervix. A syringe was filled with the relevant volume of Lugol’s iodine, then attached via a Luer-lock to the atomizer design of interest, which was held in place by a custom fixture to ensure it was centered. The fluid was then sprayed with high force along the optical axis 48 mm from the cervix. This working distance was estimated as the approximate distance the atomizer would be positioned when integrated into the Callascope’s tip. The optical axis was used for initial testing while optimizing other variables, then the integration angle was selected based on the CAD model once the design was frozen. Use of Lugol’s Iodine enables visibility of the contrast on the mock-cervix as the cloth turns brown when sprayed. Images were then analyzed with the FIJI macro to quantify coverage of the cervix. Any pooling of fluid, backsplash onto the camera, or impingement on camera FOV was also noted during testing (n = 6). The area percentage of the mock-cervix image that was saturated with contrast agent and the circularity of the area were measured with the open-source FIJI⁴⁴ software package.

Participant Enrollment

The protocol, consent process, and data storage system for the study were approved by the Duke University Medical Center (DUMC) Institutional Review Board (Pro00008173; ClinicalTrials.gov Identifier NCT00900575, date of registration May 11th, 2009). All methods were performed in accordance with the relevant guidelines and regulations. Patients at DUMC Gynecology Clinic in Durham, NC, USA scheduled to undergo a routine follow-up pelvic exam in surveillance for cervical dysplasia, visual exam of the cervix, or colposcopy were asked to participate in the study. Participants were eligible if they were female patients between 21 and 65 years of age (cervical cancer screening age) and English speakers⁴⁹. Participants outside the designated age range were excluded, as were patients in the 2nd or 3rd trimester of pregnancy. All participants provided written informed consent prior to participation.

Disinfection safety procedures

Since the Callascope comes in direct contact with mucosal membranes and blood, it underwent high-level disinfection between participants. First, it is wiped down with germicidal wipes and water to remove any gross contamination. The tip of the Callascope is then removed from the body of the device. Disinfection was performed by soaking the Callascope and its tip in an enzymatic detergent (Enzol⁵⁰) for 2 min. The Callascope and tip were then rinsed in sterile water before high level disinfection with CIDEX OPA. The device components were submerged in CIDEX OPA for 12 min, then rinsed with sterile water and wiped with lint-free cloth. This disinfection protocol is the same as that cleared by the FDA for use with a sister technology called the Pocket colposcope⁵¹.

Clinical investigation overview

Speculum-free imaging

After informed consent, a pre-exam questionnaire was administered to gather participant preferences regarding cervical screening. Participants underwent a standard Pap smear with a speculum, after which the speculum was removed. The provider then inserted the Callascope, manipulated it to position the cervix into the field of view (FOV), centered the os, and captured an image. The Callascope was removed, and post-exam surveys were completed by both participants and physicians to investigate the comfort, ease of visualization, effectiveness of manipulation, and image quality of the device (Fig. 2).

Contrast application with the atomizer

The ability of the spray mechanism to provide adequate coverage of the cervix was tested in participants undergoing colposcopy (Arm Two; Fig. 2). Contrast study participants in both groups (through the speculum and speculum-free) underwent a similar recruitment process to the speculum-free imaging-only participants.

In the first group where the Callascope was inserted through the speculum, colposcopy began as normal by inserting the speculum. The Callascope was then inserted through the speculum, the os centered, an image was captured with the Callascope. The atomizer was then used to spray acetic acid onto the cervix, after which another image was captured with the Callascope. After removing the Callascope, an image was captured with the standard-of-care digital colposcope (SOC). In a subset of eight participants, the spray and imaging processes were repeated with Lugol’s iodine after a national Lugol’s shortage was resolved³⁵. If deemed necessary based on the standard of care, biopsies were collected. The final diagnosis and histology results for each participant were collected post-hoc. In a subset of patients, videos were captured after the os was centered with the Callascope; video recorded the acetic acid application with the atomizer and at least 60 s post-acetic acid application.

At a later date, providers (2 residents and 1 attending) annotated deidentified participant images (n = 18) to assess lesion visibility and diagnostic utility. Through a Qualtrics questionnaire, physicians were asked to assess any factors impacting image quality, provide the clock face locations that they would have biopsied, and determine if there was overall sufficient visualization and image quality to interpret the image. They also provided a PowerPoint in which they marked the locations they would have biopsied on the collected images. The clock face locations were used to measure agreement; agreement in annotation and biopsy site was defined as a marked annotation within one clock face of a collected biopsy to account for differences in clock face demarcation between physicians. Agreement was calculated per participant and stratified by histology outcome. The YOLO blur detection algorithm developed by our group was used on all post-acetic acid Callascope and SOC colposcope images³⁶. Lugol’s iodine images were visually annotated for blur. Finally, to compare cervical quadrant visibility between the Callascope and standard colposcope in the through-the-speculum contrast arm, a provider reviewed each post-contrast image and recorded the number of quadrants in which the transformation zone was visible (0–4).

In the second group of contrast study participants where the Callascope was inserted without the speculum, the design of the device was changed slightly for patient comfort; the tubing for the atomizer was moved to the inside of the Callascope body. Therefore, while eleven total participants were enrolled in this arm of the study, four patients were enrolled during device development to assess the design modifications and were excluded from the results section. An additional four participants were excluded for significant unexpected mucus occlusions that prohibited contrast application. The remaining four participants underwent speculum-free insertion of the Callascope, and the cervix was manipulated into the field of view, after which video recording began. Acetic acid (n = 3) or Lugol’s iodine (n = 1) was applied using the integrated atomizer, and video capture continued for at least 60 s post-application. Frames were later extracted at baseline and at various intervals to assess visual changes over time. After recording, the Callascope was removed, and standard colposcope imaging and biopsies were performed as required.

Across all images and videos in cases where a biopsy was taken, images were manually cropped to the cervix. Series of frames taken from videos were co-registered to account for small shifts in the device over the time-series^44,52. To quantify regional tissue changes, regions of interest were annotated on the pre- and post-acetic acid images. One region of interest was selected where the biopsy was collected in clinic, and the other was positioned on a visually normal adjacent section of squamous epithelium. The same regions were selected manually in the pre- and post- images. The intensity was measured as the mean of the red, green, and blue channel intensities of all pixels in the selected region. Pixel intensity within each region of interest was measured at both timepoints, and the percent change in brightness was calculated relative to the pre-acetic acid image.

Statistical analysis

Speculum-free imaging

Pre-exam Likert scale responses on willingness to be screened for cervical cancer in different scenarios were analyzed using a non-parametric Wilcoxon signed-rank test (two-tailed, paired, with an alpha level of 0.05). Rank correlation between participants’ individual willingness to be screened with one device compared to another was determined by calculating the Spearman’s rho. In the post-exam survey, rank correlation between participants’ pain scores at each phase of the exam was also analyzed by calculating Spearman’s rho. Survey data comparing overall perception of the Callascope exam vs. the speculum exam was analyzed with a non-parametric Wilcoxon signed-rank test (two-tailed, one-sample, with an alpha level of 0.05) to compare to a null hypothesis that there was no preference between exams. Descriptive statistics were also used to summarize image quality metrics and observations about the images collected.

Power calculations conducted with the G*Power program⁵³ indicated that a paired, two-tailed Wilcoxon signed-rank test would achieve 80% power to detect effect sizes of 0.4, 0.6, 0.7, and 0.8 for sample sizes of 54, 25, 19 and 15 subjects, respectively, at a 0.05 significance level.

Contrast application in colposcopy participants

For the second arm of the study, the agreement between physician annotations of images and the locations where biopsies were obtained during the exam was reported on with descriptive statistics. The proportion of blurry versus clear images captured by the Callascope and standard-of-care (SOC) colposcope following contrast application was compared using a chi-square test. A p value of <0.05 was considered statistically significant. Additionally, to assess inter-device agreement in image clarity classification (blurry vs. clear) at the patient level, we calculated Cohen’s kappa.

Quadrant visibility was assessed for each device with either acetic acid or Lugol’s iodine and reported as the proportion of images achieving full four-quadrant, three-quadrant, or fewer than three-quadrant visualization. For paired images where both Callascope and SOC colposcope data were available for the same participant, a Wilcoxon signed-rank test was used to compare quadrant scores between devices across all paired observations.

To compare intensity metrics between the paired regions of interest, we first calculated the difference (Biopsy – Normal) for each patient for both metrics, raw intensity in the post-AA image and the percent change in intensity using both the pre- and post-contrast images. Assumptions for the paired t-test were evaluated; normality of the differences was assessed using the Lilliefors test, and the absence of extreme outliers was verified by visual inspection of histograms and boxplots. When the normality assumption was satisfied, a paired t-test was performed. If the normality assumption was violated, the non-parametric Wilcoxon signed-rank test was used as an alternative. Two-tailed tests were applied in all cases, and statistical significance was defined as p < 0.05.

Author contributions

K.E. and E.S. designed the device, performed testing, and collected data. K.E. analyzed and interpreted results and drafted the original manuscript. M.J.H. and N.K. provided clinical resources for clinical studies. M.J.H., N.R., K.E., and E.S. contributed to the study design and conceptualization. M.R. contributed to device design. N.R. provided overall scientific guidance. All authors read and approved of the final manuscript.

Data availability

The images generated and/or analyzed during the current study are not publicly available. The participant images and pathology labels used in this study are restricted by the Institutional Review Board to protect participant privacy. All other relevant data are included in the publication and supplementary materials.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s44294-026-00148-5.