Increased detection of precancerous cervical lesions with adjunctive dynamic spectral imaging

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.

Agreement on Lesion Presence and Location at Colposcopy

OBJECTIVES/PURPOSE

The reproducibility and sensitivity of image-based colposcopy is low, but agreement on lesion presence and location remains to be explored. Here, we investigate the interobserver agreement on lesions on colposcopic images by evaluating and comparing marked lesions on digitized colposcopic images between colposcopists.

METHODS

Five colposcopists reviewed images from 268 colposcopic examinations. Cases were selected based on histologic diagnosis, i.e., normal/cervical intraepithelial neoplasia (CIN)1 ( n = 50), CIN2 ( n = 50), CIN3 ( n = 100), adenocarcinoma in situ ( n = 53), and cancer ( n = 15). We obtained digitized time-series images every 7-10 seconds from before acetic acid application to 2 minutes after application. Colposcopists were instructed to digitally annotate all areas with acetowhitening or suspect of lesions. To estimate the agreement on lesion presence and location, we assessed the proportion of images with annotations and the proportion of images with overlapping annotated area by at least 4 (4+) colposcopists, respectively.

RESULTS

We included images from 241 examinations (1 image from each) with adequate annotations. The proportion with a least 1 lesion annotated by 4+ colposcopists increased by severity of histologic diagnosis. Among the CIN3 cases, 84% had at least 1 lesion annotated by 4+ colposcopists, whereas 54% of normal/CIN1 cases had a lesion annotated. Notably, the proportion was 70% for adenocarcinoma in situ and 71% for cancer. Regarding lesion location, there was no linear association with severity of histologic diagnosis.

CONCLUSION

Despite that 80% of the CIN2 and CIN3 cases were annotated by 4+ colposcopists, we did not find increasing agreement on lesion location with histology severity. This underlines the subjective nature of colposcopy.

Analysis of New Colposcopy Techniques in the Diagnosis and Evolution of SIL/CIN: Comparison of Colposcopy with the DSI System (COLPO-DSI Study)

Compared with conventional colposcopy, colposcopy assisted by DSI-map increases the detection of HSIL/CIN2+ and might help to identify the lesions more likely to regress.

INTRODUCTION

Comparison of the performance of colposcopy assisted by dynamic spectral imaging (C-DSI) with that of conventional colposcopy (CC) in the diagnosis of cervical intraepithelial neoplasia (HSIL/CIN2 or CIN3).

MATERIALS AND METHODS

A total of 1655 women were referred for colposcopy between 2012 and 2020 and included in the study. Of that total, 973 were examined by the same colposcopist with C-DSI, and 682 with CC. Comparisons between CC and C-DSI were made by using the histological diagnosis performed with a punch biopsy or loop electrosurgical excision procedure (LEEP) as the gold standard. A follow-up study was conducted until 2021 to detect progression to HSIL/CIN2 at 6, 12 and 24 months after first examination.

RESULTS

C-DSI provided higher sensitivity for the diagnosis of HSIL/CIN2 or CIN 3 than CC (sensitivity of 76.8% and 86.6% vs. 54.2% and 72.2%, respectively). In negative or ASCUS/LSIL Pap smear results, C-DSI showed higher sensitivity than CC (sensitivity of 66.7% and 61.5% vs. 21.4% and 33.3%, respectively). In contrast, these differences were not observed in high-grade Pap smears. The sensitivity of C-DSI in cases with HPV16/18 infection was stronger than that of CC (73.53% vs. 56.67%). The sensitivity of C-DSI to detect the progression to HSIL/CIN2+ during follow-up was 30, 17.6 and 35.7% at 6, 12 and 24 months, respectively.

CONCLUSIONS

The present study shows that C-DSI in women referred for colposcopy increases the HSIL/CIN 2-3 detection rate compared to conventional colposcopy. Nevertheless, C-DSI does not seem to be an important tool to predict the evolution of the lesions during follow-up.

Comparison of accuracy and reproducibility of colposcopic impression based on a single image versus a two-minute time series of colposcopic images

OBJECTIVE

Colposcopy is an important part of cervical screening/management programs. Colposcopic appearance is often classified, for teaching and telemedicine, based on static images that do not reveal the dynamics of acetowhitening. We compared the accuracy and reproducibility of colposcopic impression based on a single image at one minute after application of acetic acid versus a time-series of 17 sequential images over two minutes.

METHODS

Approximately 5000 colposcopic examinations conducted with the DYSIS colposcopic system were divided into 10 random sets, each assigned to a separate expert colposcopist. Colposcopists first classified single two-dimensional images at one minute and then a time-series of 17 sequential images as 'normal,' 'indeterminate,' 'high grade,' or 'cancer'. Ratings were compared to histologic diagnoses. Additionally, 5 colposcopists reviewed a subset of 200 single images and 200 time series to estimate intra- and inter-rater reliability.

RESULTS

Of 4640 patients with adequate images, only 24.4% were correctly categorized by single image visual assessment (11% of 64 cancers; 31% of 605 CIN3; 22.4% of 558 CIN2; 23.9% of 3412 < CIN2). Individual colposcopist accuracy was low; Youden indices (sensitivity plus specificity minus one) ranged from 0.07 to 0.24. Use of the time-series increased the proportion of images classified as normal, regardless of histology. Intra-rater reliability was substantial (weighted kappa = 0.64); inter-rater reliability was fair ( weighted kappa = 0.26).

CONCLUSION

Substantial variation exists in visual assessment of colposcopic images, even when a 17-image time series showing the two-minute process of acetowhitening is presented. We are currently evaluating whether deep-learning image evaluation can assist classification.

Value of colposcopy with dynamic spectral imaging in the detection and evolution of high-grade cervical lesions

OBJECTIVE

To analyze the performance of colposcopy assisted with dynamic spectral imaging (CC-DSI) compared to conventional colposcopy (CC) in diagnosing cervical intraepithelial neoplasia 2+ (CIN2+).

METHODS

: 973 consecutive women were referred for colposcopy between 2012 and 2016 and were examined simultaneously by CC and CC-DSI; 877 were eligible. Comparisons between CC and CC-DSI were performed against the histological diagnosis performed by either punch biopsy or loop electrosurgical excision procedure (LEEP).

RESULTS

: In final histology, 494 women had no CIN, 250 had CIN1 and 133 had CIN2+. The sensitivity to identify women with CIN2+ was significantly higher for CC-DSI compared to CC for the entire group and in the subgroup of 675 women referred with ASCUS or LSIL. In women with HPV16/18 infections, the sensitivity of CC increased with the addition of DSI from 53% to 79% (p<0.001). Using a multivariant Cox regression model, CC-DSI was an independent factor for progression of the cervical lesions (HR: 2.29, 95%CI 1.07-4.90). Other predictive factors were the number of sexual partners (HR: 1.05, 95%CI: 1.01-1.09) and anal intercourse (HR: 2.45, 95%CI 1.23-5.02).

CONCLUSION

CC-DSI improves the ability to detect cervical lesions compared to CC and could help predict their potential to progress.

Unsupervised Deep Learning Registration of Uterine Cervix Sequence Images

During a colposcopic examination of the uterine cervix for cervical cancer prevention, one or more digital images are typically acquired after the application of diluted acetic acid. An alternative approach is to acquire a sequence of images at fixed intervals during an examination before and after applying acetic acid. This approach is asserted to be more informative as it can capture dynamic pixel intensity variations on the cervical epithelium during the aceto-whitening reaction. However, the resulting time sequence images may not be spatially aligned due to the movement of the cervix with respect to the imaging device. Disease prediction using automated visual evaluation (AVE) techniques using multiple images could be adversely impacted without correction for this misalignment. The challenge is that there is no registration ground truth to help train a supervised-learning-based image registration algorithm. We present a novel unsupervised registration approach to align a sequence of digital cervix color images. The proposed deep-learning-based registration network consists of three branches and processes the red, green, and blue (RGB, respectively) channels of each input color image separately using an unsupervised strategy. Each network branch consists of a convolutional neural network (CNN) unit and a spatial transform unit. To evaluate the registration performance on a dataset that has no ground truth, we propose an evaluation strategy that is based on comparing automatic cervix segmentation masks in the registered sequence and the original sequence. The compared segmentation masks are generated by a fine-tuned transformer-based object detection model (DeTr). The segmentation model achieved Dice/IoU scores of 0.917/0.870 and 0.938/0.885, which are comparable to the performance of our previous model in two datasets. By comparing our segmentation on both original and registered time sequence images, we observed an average improvement in Dice scores of 12.62% following registration. Further, our approach achieved higher Dice and IoU scores and maintained full image integrity compared to a non-deep learning registration method on the same dataset.

Dynamic Spectral Imaging Colposcopy Versus Regular Colposcopy in Women Referred With High-Grade Cytology: A Nonrandomized Prospective Study

OBJECTIVE

The aim of the study was to evaluate the sensitivity of dynamic spectral imaging (DSI) colposcopy compared with regular colposcopy for women referred with high-grade cervical cytology.

METHODS

In a prospective, nonrandomized, multicenter study, we included women referred for colposcopy at hospital gynecology clinics with high-grade cytology. Women were examined using either a regular or DSI colposcope. In both groups, colposcopists located 1 area viewed as most suspicious. In the DSI group, this was done before viewing the DSI map. Subsequently, an area was chosen based on the worst color of the DSI map, and further additional biopsies were taken. All women had 4 cervical biopsies taken, all analyzed separately. The main outcome was sensitivity to find cervical intraepithelial neoplasia grade 2 or worse (CIN2+).

RESULTS

A total of 261 women were examined using DSI colposcopy, and 156 women were examined using regular colposcopy. The sensitivity for finding CIN2+ when using the DSI technology as an adjunctive technology was found to be 82.2% (95% CI = 75.9-87.4), based on an average of 1.4 biopsies. This was corresponding in sensitivity to 2 biopsies taken using regular colposcopy (80.3%; 95% CI = 72.3-86.8). There was no difference in sensitivity for CIN+ between the groups when 3 or more biopsies were taken.

CONCLUSIONS

We found that the DSI colposcope may help direct biopsy placement; however, the improvement is based on small differences in needed biopsies and the clinical significance of this may be small. Multiple biopsies were still superior.

Can the dynamic spectral imaging (DSI) color map improve colposcopy examination for precancerous cervical lesions? A prospective evaluation of the DSI color map in a multi-biopsy clinical setting

Background

Colposcopy serves as a subjective examination of the cervix with low sensitivity to detect cervical intraepithelial dysplasia (CIN) grade 2 or worse (CIN2 +). Dynamic spectral imaging (DSI) colposcopy has been developed to provide an objective element to cervix examinations and has been proven to increase sensitivity of detecting CIN2 + . We aimed to assess the performance of the DSI color map and compared it to histological diagnoses of cervical biopsies in determining the CIN grade present.

Methods

Women were included in a consecutive, prospective manner at Randers Regional Hospital, Denmark. Women were eligible to participate if they were referred for colposcopy due to abnormal cervical smear (threshold:  ≥ ASCUS) or follow-up after previously diagnosed CIN. All women had four biopsies taken, one directed by colposcopists alone prior to viewing the DSI color map, one directed by the worst color on the respective DSI color map, and two additional biopsies. All biopsies were analyzed separately. We calculated sensitivity, specificity, positive predictive values, and negative predictive values (NPVs) with 95% confidence intervals (CIs).

Results

A total of 800 women were recruited. Of these, 529 (66.1%) were eligible for inclusion. The sensitivity of the DSI color map was found to be 48.1% (95% CI 41.1–55.1) in finding CIN grade 2 or worse (CIN2 +) when compared to the histological diagnosis of the DSI directed biopsy. This was 42.5% (95% CI 36.7–48.5) when compared to the final histological diagnosis of all four cervical biopsies and with an NPV of 53.5% (95% CI 50.5–56.5).

Conclusion

The worst color indicated by the DSI map might not consistently reflect the true grade of cervical dysplasia present. Thus, even though the DSI color map indicates low-grade changes, colposcopists should still consider taking biopsies from the area as high-grade changes might be present.

Trial registration: NCT04249856, January 31 2020 (retrospectively registered).

Using technology to harmonise treatment approaches in colposcopy in the face of a changing environment

The overall numbers of precancerous lesions are expected to fall as human papillomavirus (HPV) vaccinated women enter the cervical screening programme. Juxtaposed against an increase in referrals from the introduction of primary high-risk HPV screening, colposcopists expect to see a decreasing incidence of high-grade cervical intraepithelial neoplasia (CIN). Correct identification of lesions will become more challenging, as the prevalence of high-grade lesions becomes minimal and conventional colposcopy is subject to a lower sensitivity. In this review, we explore the scenarios where adjunct technologies could support colposcopists to manage referrals and diagnose treatable lesions with more confidence.

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+.

Accuracy of colposcopy-directed biopsy vs dynamic spectral imaging directed biopsy in correctly identifying the grade of cervical dysplasia in women undergoing conization: A methodological study

INTRODUCTION

Dynamic spectral imaging (DSI) colposcopy has previously been found to improve sensitivity of CIN2+ detection. The aim of this study was to compare the histological diagnosis of colposcopic-directed biopsies (CDB) with that of DSI-directed biopsies in women undergoing conization, using the histological diagnosis of the conization specimen as gold standard.

MATERIAL AND METHODS

Women referred for colposcopy were included in a prospective cohort study at Randers Regional Hospital, Denmark, from January 2016 to February 2019. All women had four cervical punch biopsies taken. The first biopsy was taken from the area that appeared most abnormal by conventional colposcopy (ie, CDB) and the second biopsy from the area that appeared most abnormal using the DSI map. An additional two biopsies were taken either from other visible lesions or as random biopsies. Biopsies were analyzed separately. If any biopsies revealed cervical dysplasia of such a degree that excisional treatment was recommended, the patient was referred for conization. Subsequently, we compared the histological diagnosis of CDB and DSI-directed biopsies with that of the cone biopsy.

RESULTS

A total of 573 women were enrolled, 170 of which underwent conization. In women with an adequate colposcopy and representative biopsies (n = 124) there was an overall agreement rate between the worst biopsy diagnosis (of any four) and the conization diagnosis in 95.2% (95% CI 89.8-98.2) of women. CDB diagnosis agreed with the cone diagnosis in 80.6% (95% CI 72.6-87.2) of women. DSI-directed biopsy agreed with the cone diagnosis in 83.9% (95% CI 76.2-89.9) of women. The difference in detection rate between the CDB and the DSI-directed biopsy was, however, not significant (P = .54). Taking four biopsies increases the detection rate of cervical dysplasia to 95.2%, which was a significant increase from both CDB alone (P = .0008) and DSI-directed biopsy alone (P = .0053).

CONCLUSIONS

We found no significant difference in the ability to identify the cervical dysplasia grade between CDB and DSI-directed biopsies. A higher detection rate of cervical dysplasia was achieved with four biopsies than with one CDB biopsy or one DSI-directed biopsy.

Combined dynamic spectral imaging and routine colposcopy strategy for the diagnosis of pre-cancerous cervical lesions

The present study aimed to investigate the sensitivity and specificity of combined dynamic spectral imaging (DySI) and routine colposcopy in diagnosing pre-cancerous lesions in subjects with abnormal cytological results. The retrospective study included 146 patients diagnosed with cervical lesions and atypical squamous cells of undetermined significance according to a colposcopy examination. Data from colposcopy clinics were used to evaluate performance of DySI in different histological types. The present study evaluated the clinical performance of two different referral strategies (single diagnosis or combined diagnosis) in combination with a smear test. The sensitivity of DySI and routine colposcopy for detecting severe pre-cancerous cervical disease in patients referred with a borderline histology or mild dyskaryosis was 69% [95% confidence interval (CI): 56-82%] and 43% (95% CI: 28-57%), respectively (P=0.01). However, for the HSIL group, the sensitivity of DySI and routine colposcopy was 64 and 61%, respectively. In the HSIL group, the sensitivity was as high as 85% when the DySI method was combined with routine colposcopy. When the conditions of new screening strategies were applied to remove those patients with a negligible risk, DySI had a higher sensitivity to detect severe pre-cancerous cervical disease than conventional colposcopy. In conclusion, the combined DySI and routine colposcopy strategy had a higher diagnostic sensitivity in subjects with low or high abnormalities on cytological examination, compared with either method separately. The present study suggests that the baseline colposcopy sensitivity may be enhanced with the adjunctive use of DySI, irrespective of the cytology result for referral.