Deep Learning Diagnostic Classification of Cervical Images to Augment Colposcopic Impression.

OBJECTIVE: A deep learning classifier that improves the accuracy of colposcopic impression. METHODS: Colposcopy images taken 56 seconds after acetic acid application were processed by a cervix detection algorithm to identify the cervical region. We optimized models based on the SegFormer architecture to classify each cervix as high-grade or negative/low-grade. The data were split into histologically stratified, random training, validation, and test subsets (80%-10%-10%). We replicated a 10-fold experiment to align with a prior study utilizing expert reviewer analysis of the same images. To evaluate the model's robustness across different cameras, we retrained it after dividing the dataset by camera type. Subsequently, we retrained the model on a new, histologically stratified random data split and integrated the results with patients' age and referral data to train a Gradient Boosted Tree model for final classification. Model accuracy was assessed by the receiver operating characteristic area under the curve (AUC), Youden's index (YI), sensitivity, and specificity compared to the histology. RESULTS: Out of 5,485 colposcopy images, 4,946 with histology and a visible cervix were used. The model's average performance in the 10-fold experiment was AUC = 0.75, YI = 0.37 (sensitivity = 63%, specificity = 74%), outperforming the experts' average YI of 0.16. Transferability across camera types was effective, with AUC = 0.70, YI = 0.33. Integrating image-based predictions with referral data improved outcomes to AUC = 0.81 and YI = 0.46. The use of model predictions alongside the original colposcopic impression boosted overall performance. CONCLUSIONS: Deep learning cervical image classification demonstrated robustness and outperformed experts. Further improved by including additional patient information, it shows potential for clinical utility complementing colposcopy.

Assessing Inter-Annotator Agreement for Medical Image Segmentation.

Artificial Intelligence (AI)-based medical computer vision algorithm training and evaluations depend on annotations and labeling. However, variability between expert annotators introduces noise in training data that can adversely impact the performance of AI algorithms. This study aims to assess, illustrate and interpret the inter-annotator agreement among multiple expert annotators when segmenting the same lesion(s)/abnormalities on medical images. We propose the use of three metrics for the qualitative and quantitative assessment of inter-annotator agreement: 1) use of a common agreement heatmap and a ranking agreement heatmap; 2) use of the extended Cohen's kappa and Fleiss' kappa coefficients for a quantitative evaluation and interpretation of inter-annotator reliability; and 3) use of the Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm, as a parallel step, to generate ground truth for training AI models and compute Intersection over Union (IoU), sensitivity, and specificity to assess the inter-annotator reliability and variability. Experiments are performed on two datasets, namely cervical colposcopy images from 30 patients and chest X-ray images from 336 tuberculosis (TB) patients, to demonstrate the consistency of inter-annotator reliability assessment and the importance of combining different metrics to avoid bias assessment.

Rate of detecting CIN3+ among patients with ASC-US using digital colposcopy and dynamic spectral imaging.

The present study compared two methods for the detection of severe cervical dysplasia in women with atypical squamous cells of underdetermined significance (ASC-US) cytology; digital colposcopy with adjunctive dynamic spectral imaging (DSI) and conventional colposcopy. IMPROVE-COLPO was a two-arm cross-sectional study of US community-based colposcopy. The active (prospective) arm of this study recruited patients examined by digital colposcopy and adjunctive DSI. Preceding consecutive patients that had been examined with conventional methods were used as historical controls in the retrospective arm of the study after being matched in number to those in the prospective arm by a colposcopist. In the present study, the primary measure was the number of women detected with cervical intraepithelial neoplasia (CIN) grade 3 or worse (CIN3+) following punch biopsy. The study included 1,353 retrospective and 1,226 prospective patients eligible for this analysis who were examined by 146 colposcopists in 42 community-based clinics. The patient baseline characteristics were comparable between the two arms. The average number of biopsies taken per patient was higher among the prospective arm patients (including standard and DSI-assisted biopsies) compared with the retrospective arm control patients (1.21 vs. 0.97 respectively). Biopsy detected 31 patients with CIN3+ [2.29%; 95% confidence interval (CI), 1.56-3.24] in the retrospective arm, and 48 patients with CIN3+ (3.92%; 95% CI, 2.90-5.16) in the prospective arm. The difference in the number of patients detected with CIN3+ in the two arms of the study was 1.62% (95% CI, 0.30-3.04; P=0.022), which corresponds to a 70.9% relative increase in the prospective compared with the retrospective arm. Biopsy appeared less efficient in detecting patients with CIN3+ in the retrospective arm compared with the prospective arm. However, there was no statistically significant difference between the retrospective arm and the prospective arm in terms of: i) Biopsies taken (over the entire population) per patient detected with CIN3+ (42.2 in the retrospective arm vs. 30.8 in the prospective arm; P=0.164) and ii) positive predictive value of using biopsies to identify patients with CIN3+ (2.83 vs. 3.92; P=0.118). Adoption of digital colposcopy with DSI increased the number of biopsies collected from ASC-US patients compared with retrospective controls of standard colposcopy and detected a significantly higher number of patients who were CIN3+. The number of additional biopsies taken in the prospective arm compared with the retrospective arm was too small to explain the increased detection of patients with CIN3+ observed in the prospective arm, suggesting that biopsies in the prospective arm were better at identifying CIN3+.

Observed Colposcopy Practice in US Community-Based Clinics: The Retrospective Control Arm of the IMPROVE-COLPO Study.

OBJECTIVE: The aim of the study was to characterize colposcopy practice and management of women with cervical abnormalities in US community-based clinics. MATERIALS AND METHODS: IMPROVE-COLPO was a 2-arm study of colposcopy patients with an abnormal screening result. The prospective arm recruited women to undergo examination with a commercial digital colposcope. The retrospective-control arm collected data (chart review) from previous colposcopies performed using standard equipment and methods. From the retrospective arm, we analyzed referral trends, colposcopy and biopsy practice, and management patterns. RESULTS: We collected data of 3,602 eligible women (median age = 34 years) that had been examined from 2012 to 2017 by 154 colposcopists at 44 clinics across 12 states. Most patients were premenopausal (87.9%), privately insured (88.2%), and had a low-grade (low-grade squamous intraepithelial lesion/atypical squamous cells of undetermined significance/human papillomavirus positive) indication (87.2%). Most colposcopists performed less than 3 colposcopies monthly and their biopsy rate was 1.47 biopsies/patient for high-grade referrals and 0.97 for low-grade referrals (p < .001). Random biopsy was rare (0.4% of biopsies). Most women (74.9%) underwent endocervical sampling, including 62.5% of women aged 21 to 24 years. Colposcopic impression was frequently not reported (58.8%), and its sensitivity to predict histology-confirmed cervical intraepithelial neoplasia (CIN) 2+ as "high-grade" was 56.5% for high-grade referrals and 23.2% for low-grade referrals. Excisions often (44.5%) returned

Evaluating Colposcopy with Dynamic Spectral Imaging During Routine Practice at Five Colposcopy Clinics in Wales: Clinical Performance.

BACKGROUND: The diagnostic accuracy of colposcopy is poor for detecting precancerous cervical lesions. OBJECTIVES: We assessed the performance of colposcopy for identifying cervical intraepithelial neoplasia grade 2 or worse (CIN2+), before and after including a dynamic spectral imaging (DSI) map that quantifies and maps acetowhitening to assist subsequent biopsy of suspicious lesions. METHODS: Four hundred and twenty-five women were examined at a multi-center setting in Wales, of which 393 women were included in the final analysis. RESULTS: For all referrals, the sensitivity of conventional colposcopy for histologically confirmed CIN2+ was 51.5%, the specificity was 92.0%, the positive predictive value was 56.7%, and the negative predictive value (NPV) was 90.4%. With the incorporation of the DSI map in predicting CIN2+, these became 84.8, 61.5, 30.8, and 95.3% respectively. The increase of sensitivity was statistically significant (p < 0.001). For the 236 women having colposcopy after low-grade (LG) cytology, with the incorporation of DSI, the sensitivity for CIN2+ increased from 27.3 to 86.4% (p < 0.001) and the NPV from 92.6 to 97.8%. CONCLUSIONS: Colposcopy with DSI results in improved sensitivity to detect CIN2+ and maintains a high NPV for all referrals and especially for those with LG referral cytology.

Diagnostic accuracy of colposcopy with dynamic spectral imaging for cytology-negative/high-risk HPV positive (failed test of cure) after large loop excision of the transformation zone (LLETZ) of the cervix: Results of the DySIS colposcopy 1 study.

After treatment for cervical intraepithelial neoplasia (CIN), in the UK women who are cytology-negative, high-risk (HR) human papilloma virus (HPV) positive are referred to colposcopy. This pilot study assessed the incidence of residual/recurrent CIN and the diagnostic accuracy of colposcopy with dynamic spectral imaging (DSI) mapping in their detection.This was a prospective service evaluation carried out in a UK National Health Service (NHS) colposcopy clinic. All women, referred with negative cytology/HR-HPV positive result following treatment for CIN from March 2013 until November 2014, who were examined with the DSI digital colposcope were included. We excluded 3 cases because of poor-quality imaging from user errors. Everyday clinical practice was followed. Initial colposcopic impression, DSI map indication, and biopsy site selections were recorded. CIN2+ was considered the primary outcome and CIN of any grade a secondary outcome.A total of 105 women were included of which 5 (4.8%) had CIN2+ histology and 24 (22.9%) had CIN1. Pre-DSI map colposcopy suggested normal/low grade in all 5 of the CIN2+ cases and DSI suggested high-grade (HG) CIN in 4 of the 5 cases. Sensitivity of standard colposcopy for CIN2+ was 0%, improving to 80% with the incorporation of the DSI map.The CIN burden in this population is higher than previously expected. Colposcopic identification of HG CIN appears to improve significantly with DSI in this cohort leading to refinement in patient management. A larger, multicentric prospective study (DySIS colposcopy 2) is planned to confirm these initial findings.

Digital Colposcopy With Dynamic Spectral Imaging for Detection of Cervical Intraepithelial Neoplasia 2+ in Low-Grade Referrals: The IMPROVE-COLPO Study.

OBJECTIVE: The aim of the study was to determine, in a wide "real-world" setting, whether digital colposcopy with adjunctive dynamic spectral imaging (DSI) mapping increases the detection of women with high-grade cervical intraepithelial neoplasia (CIN). MATERIALS AND METHODS: A multicenter, two-arm, observational, cross-sectional study that recruited women 21 years and older, having colposcopy after a low-grade abnormality screening result. The prospective arm collected outcomes of digital colposcopy with DSI used for identifying biopsy sites at the colposcopists' discretion. The retrospective control arm (number of subjects matched 1:1 per colposcopist) collected outcomes of standard colposcopy. The primary outcome was histopathological detection of women with CIN 2+ by colposcopic biopsy. RESULTS: The study included 1,788 women in the retrospective and 1,857 in the prospective arm from 39 US community-based clinics. Subject characteristics were comparable. A total of 71.6% of the women in the retrospective and 71.5% in the prospective arm underwent biopsy. The average number of biopsies increased from 1.032 (retrospective) to 1.256 (prospective). The yield of CIN 2+ patients was 7.21% in the retrospective and 9.48% in the prospective arm, a 2.27% difference (95% confidence interval = 0.47%-4.07%, p = .014) and 31.4% relative increase. The yield of CIN 3+ patients was 2.07% in the retrospective and 3.23% in the prospective arm, a 1.16% (95% confidence interval = 0.12%-2.24%, p = .031) absolute difference and 56.1% relative increase. The false-positive rates for biopsied patients were comparable (64.43% vs 62.04%, p = .139). CONCLUSIONS: Digital colposcopy with the adjunctive DSI increased CIN 2+ and CIN 3+ detection in low-grade referrals compared with standard colposcopy, with a similar number of women undergoing biopsy.

Increased detection of precancerous cervical lesions with adjunctive dynamic spectral imaging.

OBJECTIVE: To validate, in US community-based colposcopy clinics, previous reports of increased detection of high-grade cervical intraepithelial neoplasia (CIN2+) with biopsies selected using dynamic spectral imaging (DSI) mapping after standard colposcopy. STUDY DESIGN: Cross-sectional observational study of 26 colposcopists across nine clinics recruiting consecutive colposcopy patients. Standard assessment with biopsy selections was completed before seeing the DSI map which was subsequently interpreted and used for additional biopsies per clinical judgment. Primary measure was the number of women with CIN2+ detected by DSI-assisted biopsies, over those detected by standard colposcopy biopsies. RESULTS: A total of 887 women were recruited. After exclusions, 881 women and 1,189 biopsies were analyzed. Standard biopsy detected 78 women with CIN2+ and DSI-assisted biopsies another 34, increasing the detection rate from 8.85% to 12.71% (p=0.00016). This was achieved with 16.16% of DSI-assisted biopsies finding CIN2+ compared to 13.24% for the preceding standard biopsies. For secondary specificity analysis, 431 women had only

Standardized Digital Colposcopy with Dynamic Spectral Imaging for Conservative Patient Management.

BACKGROUND: Colposcopy is subjective and management of young patients with high-grade disease is challenging, as treatments may impair subsequent pregnancies and adversely affect obstetric outcomes. Conservative management of selected patients is becoming more popular amongst clinicians; however it requires accurate assessment and documentation. Novel adjunctive technologies for colposcopy could improve patient care and help individualize management decisions by introducing standardization, increasing sensitivity, and improving documentation. CASE: A nulliparous 27-year-old woman planning pregnancy underwent colposcopy following high-grade cytology. The colposcopic impression was of low-grade changes, whilst the Dynamic Spectral Imaging (DSI) map of the cervix suggested potential high-grade. A DSI-directed biopsy confirmed CIN2. At follow-up, both colposcopy and DSI were suggestive of low-grade disease only, and image comparison confirmed the absence of previously present acetowhite epithelium areas. Histology of the transformation zone following excisional treatment, as per patient's choice, showed no high-grade changes. CONCLUSION: Digital colposcopy with DSI mapping helps standardize colposcopic examinations, increase diagnostic accuracy, and monitor cervical changes over time, improving patient care. When used for longitudinal tracking of disease and when it confirms a negative colposcopy, it can help towards avoiding overtreatment and hence decrease morbidity related to cervical excision.

How Colposcopy Misses Invasive Cervical Cancer: A Case Report from the IMPROVE-COLPO Study.

Colposcopy with biopsy is pivotal to cervical cancer prevention programs as it is called to identify the precancerous lesions on patients that screen positive. We present a cervical cancer case from IMPROVE-COLPO, a postmarketing, multicenter, two-arm observational study on US community colposcopy that collects outcomes with the adjunctive Dynamic Spectral Imaging (DSI) in its prospective arm. A 45-year-old woman was seen for suffering of heavy periods. She had cytology of Atypical Squamous Cells of Undetermined Significance (ASC-US) and was Human Papillomavirus (HPV) positive. Her colposcopy did not recognize the underlying condition and opted for no biopsy. The DSI assessment led to a biopsy of a lesion challenging traditional colposcopic templates: small, away from the cervical os, with slow acetowhitening development. Pathology review revealed the presence of invasive squamous carcinoma. In the era of sensitive screening, it is concerning that invasive cancers can challenge colposcopy and that the way to improve colposcopy is to collect multiple biopsies from each patient. The case presented indicates that the adjunctive objective assessment by DSI increases reassurance that observations outside of traditional standard visual templates are not underestimated or ignored.

Controlled disorder in plant light-harvesting complex II explains its photoprotective role.

The light-harvesting antenna of photosystem II (PSII) has the ability to switch rapidly between a state of efficient light use and one in which excess excitation energy is harmlessly dissipated as heat, a process known as qE. We investigated the single-molecule fluorescence intermittency of the main component of the PSII antenna (LHCII) under conditions that mimic efficient use of light or qE, and we demonstrate that weakly fluorescing states are stabilized under qE conditions. Thus, we propose that qE is explained by biological control over the intrinsic dynamic disorder in the complex-the frequencies of switching establish whether the population of complexes is unquenched or quenched. Furthermore, the quenched states were accompanied by two distinct spectral signatures, suggesting more than one mechanism for energy dissipation in LHCII.

Excited-state dynamics of carotenoids in light-harvesting complexes. 1. Exploring the relationship between the S1 and S* states.

Dispersed transient absorption spectra collected at variable excitation intensities in combination with time-resolved signals were used to explore the underlying connectivity of the electronic excited-state manifold of the carotenoid rhodopin glucoside in the light-harvesting 2 complex isolated from Rhodopseudomonas acidophila. We find that the S state, which was recently identified as an excited state in carotenoids bound in bacterial light-harvesting complexes, exhibits a different response to the increase of excitation intensity than the S(1) state, which suggests that the models used so far to describe the excited states of carotenoids are incomplete. We propose two new models that can describe both the time-resolved and the intensity-dependent data; the first postulates that S(1) and S* are not populated in parallel after the decay of the initially excited S(2) state but instead result from the excitation of distinct ground-state subpopulations. The second model introduces a resonantly enhanced light-induced transition during excitation, which promotes population to higher-lying excited states that favors the formation of S* over S(1). Multiwavelength target analysis of the time-resolved and excitation-intensity dependence measurements were used to characterize the involved states and their responses. We show that both proposed models adequately fit the measured data, although it is not possible to determine which model is most apt. The physical origins and implications of both models are explored.

Excited-state dynamics of carotenoids in light-harvesting complexes. 2. Dissecting pulse structures from optimal control experiments.

Dispersed multipump-probe (PPP) spectroscopy was used to explore the role of saturation, annihilation, and structured pulses in recent coherent control experiments on the light-harvesting 2 complex from Rhodopseudomonas acidophila (Herek et al. Nature 2002, 417, 533). We discuss the complimentary aspects between the PPP technique and coherent control studies, in particular the ability to dissect complicated pulse structures and the utility in exploring incoherent mechanisms. With the aid of a simple multistate model involving only population dynamics, we illustrate how the optimized structured pulses may be explained in terms of an interplay between excited-state populations, saturation, and annihilation. Furthermore, we discuss the experimental conditions that are required for incoherent effects to contribute to control experimental signals, with particular emphasis on pulse intensities, and show that the optimization of a ratio of conservative signals (i.e., not modulated by external dynamics) is required to exclude saturation effects from coherent control studies.

Use of ultrafast dispersed pump-dump-probe and pump-repump-probe spectroscopies to explore the light-induced dynamics of peridinin in solution.

Optical pump-induced dynamics of the highly asymmetric carotenoid peridinin in methanol was studied by dispersed pump-probe, pump-dump-probe, and pump-repump-probe transient absorption spectroscopy in the visible region. Dispersed pump-probe measurements show that the decay of the initially excited S2 state populates two excited states, the S1 and the intramolecular charge-transfer (ICT) state, at a ratio determined by the excitation wavelength. The ensuing spectral evolution occurs on the time scale of a few picoseconds and suggests the equilibration of these states. Dumping the stimulated emission of the ICT state with an additional 800-nm pulse after 400- and 530-nm excitation preferentially removes the ICT state contribution from the broad excited-state absorption, allowing for its spectral characterization. At the same time, an unrelaxed ground-state species, which has a subpicosecond lifetime, is populated. The application of the 800-nm pulse at early times, when the S2 state is still populated, led to direct generation of the peridinin cation, observed for the first time in a transient absorption experiment. The excited and ground electronic states manifold of peridinin has been reconstructed using target analysis; this approach combined with the measured multipulse spectroscopic data allows us to estimate the spectra and time scales of the corresponding transient states.

Spectroscopic characterization of the excitation energy transfer in the fucoxanthin-chlorophyll protein of diatoms.

We characterized the energy transfer pathways in the fucoxanthin-chlorophyll protein (FCP) complex of the diatom Cyclotella meneghiniana by conducting ultrafast transient absorption measurements. This light harvesting antenna has a distinct pigment composition and binds chlorophyll a (Chl-a), fucoxanthin and chlorophyll c (Chl-c) molecules in a 4:4:1 ratio. We find that upon excitation of fucoxanthin to its S2 state, a significant amount of excitation energy is transferred rapidly to Chl-a. The ensuing dynamics illustrate the presence of a complex energy transfer network that also involves energy transfer from the unrelaxed or 'hot' intermediates. Chl-c to Chl-a energy transfer occurs on a timescale of a 100 fs. We observe no significant spectral evolution in the Chl-a region of the spectrum. We have applied global and target analysis to model the measured excited state dynamics and estimate the spectra of the states involved; the energy transfer network is discussed in relation to the pigment organization of the FCP complex.

The charge-transfer character of the S0 --> S2 transition in the carotenoid peridinin is revealed by Stark spectroscopy.

Peridinin, the carotenoid in the peridinin chlorophyll a protein (PCP), was studied by Stark (electroabsorption) spectroscopy to determine the change in electrostatic properties produced on excitation within the absorption band, in methyl tetrahydrofuran (MeTHF) versus ethylene glycol (EG), at 77 K. Strikingly, a large change in the permanent dipole moment (|Deltamu|) was found between the ground state, S(0) (1(1)A(g)(-)), and the Franck-Condon region of the S(2) (1(1)B(u)(+)) excited state, in both MeTHF (22 D) and EG (approximately 27 D), thus revealing the previously unknown charge transfer (CT) character of this pi-pi transition in peridinin. Such a large |Deltamu| produced on excitation, we suggest, facilitates the bending of the lactone moiety, toward which charge transfer occurs, and the subsequent formation of the previously identified intramolecular CT (ICT) state at lower energy. This unexpectedly large S(2) dipole moment, which has not been predicted even from high-level electronic structure calculations, is supported by calculating the shift of the peridinin absorption band as a function of solvent polarity, using the experimentally derived result. Overall, the photoinduced charge transfer uncovered here is expected to affect the excited-state reactivity of peridinin and, within the protein, be important for efficient energy transfer from the carotenoid S(2) and S(1)/ICT states to the chlorophylls in PCP.

Resolving the excited state equilibrium of peridinin in solution.

The carotenoid peridinin is abundant in the biosphere, as it is the main pigment bound by the light-harvesting complexes of dinoflagellates, where it collects blue and green sunlight and transfers energy to chlorophyll a with high efficiency. Its molecular structure is particularly complex, giving rise to an intricate excited state manifold, which includes a state with charge-transfer character. To disentangle the excited states of peridinin and understand their function in vivo, we applied dispersed pump-probe and pump-dump-probe spectroscopy. The preferential depletion of population from the intramolecular charge transfer state by the dump pulse demonstrates that the S(1) and this charge transfer state are distinct entities. The ensuing dump-induced dynamics illustrates the equilibration of the two states which occurs on the time scale of a few picoseconds. Additionally, the dump pulse populates a short-lived ground state intermediate, which is suggestive of a complex relaxation pathway, probably including structural reorientation or solvation of the ground state. These findings indicate that the unique intramolecular charge transfer state of peridinin is an efficient energy donor to chlorophyll a in the peridinin-chlorophyll-protein complex and thus plays a significant role in global light harvesting.

An alternative carotenoid-to-bacteriochlorophyll energy transfer pathway in photosynthetic light harvesting.

Blue and green sunlight become available for photosynthetic energy conversion through the light-harvesting (LH) function of carotenoids, which involves transfer of carotenoid singlet excited states to nearby (bacterio)chlorophylls (BChls). The excited-state manifold of carotenoids usually is described in terms of two singlet states, S(1) and S(2), of which only the latter can be populated from the ground state by the absorption of one photon. Both states are capable of energy transfer to (B)Chl. We recently showed that in the LH1 complex of the purple bacterium Rhodospirillum rubrum, which is rather inefficient in carotenoid-to-BChl energy transfer, a third additional carotenoid excited singlet state is formed. This state, which we termed S*, was found to be a precursor on an ultrafast fission reaction pathway to carotenoid triplet state formation. Here we present evidence that S* is formed with significant yield in the LH2 complex of Rhodobacter sphaeroides, which has a highly efficient carotenoid LH function. We demonstrate that S* is actively involved in the energy transfer process to BChl and thus have uncovered an alternative pathway of carotenoid-to-BChl energy transfer. In competition with energy transfer to BChl, fission occurs from S*, leading to ultrafast formation of carotenoid triplets. Analysis in terms of a kinetic model indicates that energy transfer through S* accounts for 10-15% of the total energy transfer to BChl, and that inclusion of this pathway is necessary to obtain a highly efficient LH function of carotenoids.