Background: Cervical cancer disproportionally harms women in low and middle income countries (LMICs). There is increasing interest in automated visual evaluation (AVE) – using artificial intelligence to analyze cervical images at the point of care (PoC) – for managing patients in LMICs. AVE has a diagnostic component (for pathology) and a quality component (to ensure image adequacy). The quality component must run on the imaging device at the PoC, and is limited by its processors for computation. Methods: A novel, multiple-module algorithm for assessing cervical image quality was developed in an Android application. One module located the cervix and another sought objects obstructing the transformation zone. The cervix locator module is an object detection model that determined the bounding box of the cervix. Models trained on multiple architectures (YOLOv5 and EfficientDet-Lite2) with the same data were compared. For obstructions classification, a multi-task model was trained to detect 5 common obstructions (blood, SCJ inside of os, loose vaginal walls, mucus, blur/glare) and obstruction-free cervix. Performance of the model’s tasks were compared for 2 different imaging devices. Results and Discussion: The cervix locator performed better and was faster for YOLOv5, although differences were minimal. In the obstructions classifier, 4 different tasks (loose vaginal walls, blood, SCJ inside of os, and obstruction-free) performed satisfactorily. For all modules, the full computation time was <10 sec. Both modules met the desired performance thresholds for image adequacy assessment. The algorithm shown here is to our knowledge, the first AVE quality classifier running on a mobile device.

Cervical cancer disproportionately affects low and middle income countries. Automated visual evaluation – using deep learning to analyze a digital cervix photograph – has been proposed for patient management. Image quality remains a key challenge, as it can be degraded by many types of image defects. A series of such defects were artificially added to a test set consisting of N=344 digitized cervigram images from existing studies. Replicate test sets were created for different image defects: blur, recoloring, obstructions of different colors and directions, rotations, and white Gaussian noise. The augmented images were evaluated by a classifier. The two most significant image defects were blur and Gaussian noise.

The burden of cervical cancer disproportionately falls on low- and middle-income countries (LMICs). Automated visual evaluation (AVE) is a technology being considered as an adjunct tool for the management of HPV-positive women. AVE involves analysis of a white light illuminated cervical image using machine learning classifiers. It is of importance to analyze various impacts of different kinds of image degradations on AVE. In this paper, we report our work regarding the impact of one type of image degradation, Gaussian noise, and one of its remedies we have been exploring. The images, originated from the Natural History Study (NHS) and ASCUS-LSIL Triage Study (ALTS), were modified by the addition of white Gaussian noise at different levels. The AVE pipeline used in the experiments consists of two deep learning components: a cervix locator which uses RetinaNet (an object detection network), and a binary pathology classifier that uses the ResNeSt network. Our findings indicate that Gaussian noise, which frequently appears in low light conditions, is a key factor in degrading the AVE's performance. A blind image denoising technique which uses Variational Denoising Network (VDNet) was tested on a set of 345 digitized cervigram images (115 positives) and evaluated both visually and quantitatively. AVE performances on both the synthetically generated noisy images and the corresponding denoised images were examined and compared. In addition, the denoising technique was evaluated on several real noisy cervix images captured by a camera-based imaging device used for AVE that have no histology confirmation. The comparison between the AVE performances on images with and without denoising shows that denoising can be effective at mitigating classification performance degradation.

There is limited access to effective cervical cancer screening programs in many resource-limited settings, resulting in continued high cervical cancer burden. Human papillomavirus (HPV) testing is increasingly recognized to be the preferable primary screening approach if affordable due to superior long-term reassurance when negative and adaptability to self-sampling. Visual inspection with acetic acid (VIA) is an inexpensive but subjective and inaccurate method widely used in resource-limited settings, either for primary screening or for triage of HPV-positive individuals. A deep learning (DL)-based automated visual evaluation (AVE) of cervical images has been developed to help improve the accuracy and reproducibility of VIA as assistive technology. However, like any new clinical technology, rigorous evaluation and proof of clinical effectiveness are required before AVE is implemented widely. In the current article, we outline essential clinical and technical considerations involved in building a validated DL-based AVE tool for broad use as a clinical test.

A multi-spectral imaging system was built from low cost components: LEDs and an area-scan camera, that are all housed within a case and controlled by a tablet computer. The system can capture images of tissue at 14 different wavelengths in < 10 seconds. Spectra derived from different lateral positions in the images were then fit to a theoretical model based on GPU Monte Carlo simulations in order to estimate the scattering and absorption properties of the tissue at different layers. To better characterize the system's ability to measure changes in tissue oxy- and deoxy-hemoglobin content, images of the forearm of healthy volunteers were imaged before, during, and after short term ischemia and then reperfusion of the arm, which lowered the amount of oxyhemoglobin in the tissue. To decrease tissue oxygen saturation, blood flow to the arm was restricted for 120 sec using a sphygmomanometer (blood pressure cuff), with pressure levels of 170 mm Hg. Repeated measurements were captured with the arm held in a special mount with an aperture built to fix the tissue in place. Overall, the before, during, and after spectra, where there are notable differences between oxy and deoxy-hemoglobin. The analyses showed a significant decrease in oxygen saturation of the venous plexus layer, with moderate changes in blood content. However, changes in the error function were much more sensitive to blood content than oxygen saturation. These results suggest that changes in oxygen saturation levels can be measured using a low cost setup, although at lower accuracy relative to blood content.

Cervical cancer is a leading cause of death for women in low resource settings. Visual methods for cervical cancer screening have become more widespread. To improve diagnosis of cervical precancerous lesions, a smartphone-based mobile colposcope was developed that uses auxiliary lens and light source inside a custom-designed case. However, acquiring a sharp image in a clinical setting using the mobile colposcope is tricky. For example, trying to use the phone's auto-focus functionality struggles with the external lens placed in front of the phone's internal lens, because translation of the internal lens has a non-trivial effect the image. Moreover, auto-focus algorithm struggles with the high contrast caused by artifacts as patients' vaginal walls and pubic hair. A more robust algorithm that feeds commands back to the phone's camera module is needed. Previously, a classifier that measures image sharpness was presented. Implementing a method to correct for an out of focus image requires manipulating the smartphone's camera control parameters. This can be done either through the phone's operating system (Camera 2 API) or through the manufacturer's camera interface (Samsung Camera SDK), as called for from the application. This paper reviews how manipulations in a smartphone app affects image quality. In addition to image sharpness, analyses of brightness and color are also presented. Special apps that sweep through camera conditions were developed. Sample images from both anatomical models and calibration targets are given.

Background: When running large trials, histopathology services are used to assess the state of a tissue. However, in many clinics in low resource settings there are large variations in quality of such services, specifically in biopsy processing and histopathological interpretation/assessment of images. Quality assurance (QA) is needed, but it involves physically mailing slides to a remote clinic. A telemedicine solution can address this challenge. Methods: A novel smartphone adapter for microscopes was developed, consisting of a 3D printed attachment and software integration for the image capture. The attachment is used to couple the eyepiece of a low end microscope to a smartphone (Samsung J530). Image capture was controlled through the EVA System app. The entire system was characterized optically using standard calibration targets. Additionally, images captured on the attachment were compared to the standard method of shipping and scanning slides in a high end slice scanner at a remote clinic. Results: The resolution of the entire system (microscope + phone) with a 40X objective was < 1 mu m. The system is currently undergoing testing in Nigeria as part of a broader cervical cancer screening study.(1) Preliminary testing showed similar image quality between the smartphone-based system and high end scanner. Whole slide imaging requires stitching together images into a mosaic, made possible by a mobile application. Conclusion: The results here show that coupling a low end microscope to a smartphone yields similar results to a transporting slides to a high end microscope. Such an attachment can thus potentially provide a telemedicine solution to researchers in low resource settings.

Smartphones are currently used in many medical applications and are more frequently being integrated into medical imaging devices. The regulatory requirements in existence today however, particularly the standardization of smartphone imaging through validation and verification testing, only partially cover imaging characteristics with a smartphone. Specifically, it has been shown that smartphone camera specifications are of sufficient quality for medical imaging, and there are devices which comply with the FDA's regulatory requirements for a medical device such as a device's field of view, direction of viewing and optical resolution and optical distortion. However, these regulatory requirements do not call specifically for color testing. Images of the same object using automatic settings or different light sources can show different color composition. Experimental results showing such differences are presented. Under some circumstances, such differences in color composition could potentially lead to incorrect diagnoses. It is therefore critical to control the smartphone camera and illumination parameters properly. This paper examines different smartphone camera settings that affect image quality and color composition. To test and select the correct settings, a test methodology is proposed. It aims at evaluating and testing image color correctness and white balance settings for mobile phones and LED light sources. Emphasis is placed on color consistency and deviation from gray values, specifically by evaluating the.C values based on the CIEL*a*b* color space. Results show that such standardization minimizes differences in color composition and thus could reduce the risk of a wrong diagnosis.

Cervical cancer is the fourth most common cancer among women worldwide and is especially prevalent in low resource settings due to lack of screening and treatment options. Visual inspection with acetic acid (VIA) is a widespread and cost-effective screening method for cervical pre-cancer lesions, but accuracy depends on the experience level of the health worker. Digital cervicography, capturing images of the cervix, enables review by an off-site expert or potentially a machine learning algorithm. These reviews require images of sufficient quality. However, image quality varies greatly across users. A novel algorithm was developed to evaluate the sharpness of images captured with the MobileODT’s digital cervicography device (EVA System), in order to, eventually provide feedback to the health worker. The key challenges are that the algorithm evaluates only a single image of each cervix, it needs to be robust to the variability in cervix images and fast enough to run in real time on a mobile device, and the machine learning model needs to be small enough to fit on a mobile device’s memory, train on a small imbalanced dataset and run in real-time. In this paper, the focus scores of a preprocessed image and a Gaussian-blurred version of the image are calculated using established methods and used as features. A feature selection metric is proposed to select the top features which were then used in a random forest classifier to produce the final focus score. The resulting model, based on nine calculated focus scores, achieved significantly better accuracy than any single focus measure when tested on a holdout set of images. The area under the receiver operating characteristics curve was 0.9459.

Rhinolaryngoscopy remains difficult to perform in resource-limited settings due to the high cost of purchasing and maintaining equipment as well as the need for specialists to interpret exam findings. While the lack of expertise can be obviated by adopting telemedicine-based approaches, the capture, storage, and sharing of images/video is not a common native functionality of medical devices. Most rhinolaryngoscopy systems consist of an endoscope that interfaces with the patient’s naso/oropharynx, and a tower of modules that record video/images. However, these expensive and bulky modules can be replaced by a smartphone that can fulfill the same functions but at a lower cost. To demonstrate this, a commercially available rhinolaryngoscope was coupled to a smartphone using a 3D-printed adapter. Software developed for other clinical applications was repurposed for ENT use, including an application that controls image and video capture, a HIPAA-compliant image/video storage and transfer cloud database, and customized software features developed to improve practitioner competency. Audio recording capabilities to assess speech pathology were also integrated into the smartphone rhinolaryngoscope system. The illumination module coupled onto the endoscope remained unchanged. The spatial resolution of the rhinolaryngoscope system was defined by the fiber diameter of endoscope fiber bundle, rather than the smartphone camera. The mobile rhinolaryngoscope system was used with appropriate patients by a general practitioner in an office setting. The general practitioner then consulted with an ENT specialist via the HIPAA compliant cloud database and workflow modules on difficult cases. These results suggest the smartphone-based rhinolaryngoscope holds promise for use in low-resource settings.

Optical spectral images can be used to estimate the amount of bulk absorbers in tissues, specifically oxy- and deoxyhemoglobin, as well as scattering parameters. Most systems that capture spectral image data are large, heavy, and expensive. This paper presents a full end-to-end analysis of a low-cost reflectance-mode multispectral imaging system operating in the visible and near-infrared spectra. The system consists of 13 LEDs mounted on a printed circuit board, a monochrome machine vision camera, and a tablet computer to control the hardware. The bill of materials for the system is less than $1000. Hardware design and implementation are detailed. Calibration, image capture, and preprocessing are also discussed. In validation experiments, excellent agreement is observed in diffuse reflectance measurements between the spectral camera setup and a spectrometer. To demonstrate that such spectral image data can yield meaningful optical measurements in vivo, the forearms of eight volunteers are imaged in the system. Their data are then analyzed to estimate the tissue optical properties of different skin layers using a Monte Carlo lookup table. In three volunteers, spectral images are captured before and after inducing erythema using a warm wet towel. Across the three subjects, a clear increase in the blood content of the superficial plexus layer was observed as a result of the erythema. Collectively, these findings suggest that a low-cost system can capture accurate spectral data and that clinically meaningful information can be derived from it.

Cervical cancer is a leading cause of death for women in low resource settings. In order to better detect cervical dysplasia, a low cost multi-spectral colposcope was developed utilizing low costs LEDs and an area scan camera. The device is capable of both traditional colposcopic imaging and multi-spectral image capture. Following initial bench testing, the device was deployed to a gynecology clinic where it was used to image patients in a colposcopy setting. Both traditional colposcopic images and spectral data from patients were uploaded to a cloud server for remote analysis. Multi-spectral imaging (similar to 30 second capture) took place before any clinical procedure; the standard of care was followed thereafter. If acetic acid was used in the standard of care, a post-acetowhitening colposcopic image was also captured. In analyzing the data, normal and abnormal regions were identified in the colposcopic images by an expert clinician. Spectral data were fit to a theoretical model based on diffusion theory, yielding information on scattering and absorption parameters. Data were grouped according to clinician labeling of the tissue, as well as any additional clinical test results available (Pap, HPV, biopsy). Altogether, N=20 patients were imaged in this study, with 9 of them abnormal. In comparing normal and abnormal regions of interest from patients, substantial differences were measured in blood content, while differences in oxygen saturation parameters were more subtle. These results suggest that optical measurements made using low cost spectral imaging systems can distinguish between normal and pathological tissues.

Multispectral imaging holds great promise as a non-contact tool for the assessment of tissue composition. Performing multi - spectral imaging on a hand held mobile device would allow to bring this technology and with it knowledge to low resource settings to provide a state of the art classification of tissue health. This modality however produces considerably larger data sets than white light imaging and requires preliminary image analysis for it to be used. The data then needs to be analyzed and logged, while not requiring too much of the system resource or a long computation time and battery use by the end point device. Cloud environments were designed to allow offloading of those problems by allowing end point devices (smartphones) to offload computationally hard tasks. For this end we present a method where the a hand held device based around a smartphone captures a multi - spectral dataset in a movie file format (mp4) and compare it to other image format in size, noise and correctness. We present the cloud configuration used for segmenting images to frames where they can later be used for further analysis.

Cervical cancer disproportionately affects women living in low- and middle-income countries. To address this global crisis, many governments and NGOs have implemented community-based screening and treatment programs at outreach camps. Here, high volumes of patients are able to access care: screening and diagnosis followed by immediate treatment of precancerous lesions onsite. However, monitoring and evaluation (M&E) of these efforts presents challenges, since each event typically relies on a different health workforce, and refers patients to different facilities for follow up and advanced care. To address these challenges, a digital imaging intervention was deployed at several screening camps in East Africa. Trained nurses screened women using a connected low-cost mobile colposcope built around a smartphone. A decision support job aid was integrated into the app controlling the device, guiding nurses and recording their diagnosis and treatment decisions. Aggregating the data from the job aid allowed M&E of the screening camp in real-time. In this paper, the M&E data from 2 different screening camps in East Africa are compared. Additionally, screening camps are compared to stationary clinics. Differences in the patient screening times, treatment rates, and individual nurse statistics were all documented through the job aid allowing for much improved epidemiological information following outreach events thus enabling targeted program improvements and provider training. Reporting data from screening camps were also shared online via public web pages, facilitating broader dissemination of health needs in specific East African communities, and sparking conversations with regional stakeholders about local disease burden.

Multispectral imaging of human tissue is a powerful method that allows for quantify scattering and absorption parameters of the tissue and differentiate tissue types or identify pathology. This method requires imaging at multiple wavelengths and then fitting the measured data to a model based on light transport theory. Earlier, a mobile phone based multi-spectral imaging system was developed to image the uterine cervix from the colposcopy geometry, outside the patient's body at a distance of 200-300 mm. Such imaging of a distance object has inherent challenges, as bright and homogenous illumination is required. Several solutions addressing this problem were developed, with varied degrees of success. In this paper, several multi-spectral illumination setups were developed and tested for brightness and uniformity. All setups were specifically designed with low cost in mind, utilizing a printed circuit board with surface-mounted LEDs. The three setups include: LEDs illuminating the target directly, LEDs illuminating focused by a 3D printed miniature lens array, and LEDs coupled to a mixing lens and focusing optical system. In order to compare the illumination uniformity and intensity performance two experiments were performed. Test results are presented, and various tradeoffs between the three system configurations are discussed. Test results are presented, and various tradeoffs between the three system configurations are discussed.

The proceedings contain 10 papers. The topics discussed include: a customized scope head for OCT-guided laser stimulation of the vagus nerve;customization and testing of a mobile reader app for an open-access SARS-CoV-2 antigen lateral flow assay;color and image processing for output extraction of an LSPR sensor;automated miniscope for fluorescent cell counting applications;reverse tuning of whispering gallery mode microresonators;smartphone spectroscopy for melanoma detection;light-assisted drying (LAD) for anhydrous preservation of biologics: processing sample volumes comparable to a therapeutic dose;does the accuracy of the fingertip sphygmomanometer improve by measuring the outside temperature?;application of laser speckles and deep learning in discrimination of supermicroplastics and zooplanktons;and analysis of digital noise reduction methods on classifiers used in automated visual evaluation in cervical cancer screening.

Multi-spectral imaging systems are often expensive and bulky. An innovative multi-spectral imaging system was fitted onto a mobile colposcope, an imaging system built around a smartphone in order to image the uterine cervix from outside the body. The multi-spectral mobile colposcope (MSMC) acquires images at different wavelengths. This paper presents the clinical testing of MSMC imaging (technical validation of the MSMC system is described elsewhere(1)). Patients who were referred to colposcopy following abnormal screening test (Pap or HPV DNA test) according to the standard of care were enrolled. Multi-spectral image sets of the cervix were acquired, consisting of images from the various wavelengths. Image acquisition took 1-2 sec. Areas suspected for dysplasia under white light imaging were biopsied, according to the standard of care. Biopsied sites were recorded on a clockface map of the cervix. Following the procedure, MSMC data was processed from the sites of biopsied sites. To date, the initial histopathological results are still outstanding. Qualitatively, structures in the cervical images were sharper at lower wavelengths than higher wavelengths. Patients tolerated imaging well. The result suggests MSMC holds promise for cervical imaging.

Cervical cancer is the leading cause of cancer death for women in low resource settings, often affecting the most economically disenfranchised segment of the population. The key challenge with cervical cancer is the lack of an effective screening program for many of the at-risk, difficult-to-reach women. Outreach programs that utilize mobile clinics to increase access to screening and care in Baja California have been developed. However, many barriers such as quality assurance, efficient referral remained a challenge in this region. Visualization-based co-tests together with cytology (Pap smears) as a primary screen have been proposed. Here, the mobile colposcope of the enhanced visual assessment (EVA) is used to capture an image immediately following a Pap smear. EVA images were reviewed by expert colposcopists. Initial or preliminary data from pilot services showed that Pap false positives and Pap false negatives maybe reduced by expert review of EVA images. This suggests that reviewing of EVA images may be instrumental in catching inaccurate Pap results, thereby improving care. Thus, there is a need to further explore the benefits of using EVA as additional information when conducting Pap smear screenings.

Cervical cancer screening presents a great challenge for clinicians across the developing world. In many countries, cervical cancer screening is done by visualization with the naked eye. Simple brightfield white light imaging with photo documentation has been shown to make a significant impact on cervical cancer care. Adoption of smartphone based cervical imaging devices is increasing across Africa. However, advanced imaging technologies such as multispectral imaging systems, are seldom deployed in low resource settings, where they are needed most. To address this challenge, the optical system of a smartphone-based mobile colposcopy imaging system was refined, integrating components required for low cost, portable multi-spectral imaging of the cervix. This paper describes the refinement of the mobile colposcope to enable it to acquire images of the cervix at multiple illumination wavelengths, including modeling and laboratory testing. Wavelengths were selected to enable quantifying the main absorbers in tissue (oxyand deoxy-hemoglobin, and water), as well as scattering parameters that describe the size distribution of scatterers. The necessary hardware and software modifications are reviewed. Initial testing suggests the multi-spectral mobile device holds promise for use in low-resource settings.