Dataset Information

Deep learning on reflectance confocal microscopy improves Raman spectral diagnosis of basal cell carcinoma.

ABSTRACT:

Significance

Raman spectroscopy (RS) provides an automated approach for assisting Mohs micrographic surgery for skin cancer diagnosis; however, the specificity of RS is limited by the high spectral similarity between tumors and normal tissues structures. Reflectance confocal microscopy (RCM) provides morphological and cytological details by which many features of epidermis and hair follicles can be readily identified. Combining RS with deep-learning-aided RCM has the potential to improve the diagnostic accuracy of RS in an automated fashion, without requiring additional input from the clinician.

Aim

The aim of this study is to improve the specificity of RS for detecting basal cell carcinoma (BCC) using an artificial neural network trained on RCM images to identify false positive normal skin structures (hair follicles and epidermis).

Approach

Our approach was to build a two-step classification model. In the first step, a Raman biophysical model that was used in prior work classified BCC tumors from normal tissue structures with high sensitivity. In the second step, 191 RCM images were collected from the same site as the Raman data and served as inputs for two ResNet50 networks. The networks selected the hair structure and epidermis images, respectively, within all images corresponding to the positive predictions of the Raman biophysical model with high specificity. The specificity of the BCC biophysical model was improved by moving the Raman spectra corresponding to these selected images from false positive to true negative.

Results

Deep-learning trained on RCM images removed 52% of false positive predictions from the Raman biophysical model result while maintaining a sensitivity of 100%. The specificity was improved from 84.2% using Raman spectra alone to 92.4% by integrating Raman spectra with RCM images.

Conclusions

Combining RS with deep-learning-aided RCM imaging is a promising tool for guiding tumor resection surgery.

SUBMITTER: Chen M

PROVIDER: S-EPMC9243521 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:ImportanceThe clinical diagnosis of conjunctival and eyelid margin tumors is challenging, and new noninvasive imaging techniques could be valuable in this field.ObjectiveTo assess the diagnostic accuracy of handheld in vivo reflectance confocal microscopy (IVCM) for the diagnosis of eyelid margin and conjunctival tumors.DesignA prospective observational study was conducted at University Hospital of Saint-Etienne from January 2, 2011, to December 31, 2016 (inclusion of patients until December 31, 2015, and follow-up until December 31, 2016). A total of 278 consecutive patients with eyelid margin or conjunctival lesions were included. Conjunctival lesions were diagnosed with a conventional clinical examination using a slitlamp and by handheld IVCM. Final diagnoses were established by histopathologic examination for 155 neoformations suspicious for being malignant through clinical and/or IVCM examination that were excised and on follow-up of 12 months or longer for the remaining 140 lesions.Main outcomes and measuresSensitivity, specificity, and positive and negative predictive values for malignant tumors of the conjunctiva and eyelid margin were calculated using clinical examination with slitlamp and handheld IVCM.ResultsIn the 278 patients (136 [48.9%] females; mean [SD] age, 59 [21] years), a total of 166 eyelid margin and 129 conjunctival lesions were included in the analysis. Of the 155 excised neoformations with a histopathologic diagnosis, IVCM showed higher sensitivity compared with clinical examination conducted with the slitlamp for malignant tumors of the eyelid margin (98% vs 92%) and conjunctiva (100% vs 88%). The specificity for malignant eyelid margin tumors was higher for IVCM than for slitlamp examination (74% vs 46%), but slightly less for malignant conjunctival tumors (78% vs 88%). Analysis of all neoformations (155 excised and 140 in follow-up) confirmed these differences in the diagnostic accuracy of the clinical examination and IVCM. The presence of hyperreflective Langerhans cells mimicking malignant melanocytes was the main cause for misdiagnosis of malignant conjunctival tumors with IVCM.Conclusions and relevanceHandheld IVCM could be a useful tool for the identification of malignant conjunctival tumors. Further studies are required to confirm the usefulness of this device and identify possible features that can differentiate Langerhans cells from malignant melanocytes to prevent the misdiagnosis of melanoma using IVCM.

Project description:ImportanceBenign melanotic macules (MAC) are the most frequent cause of lip pigmentation and sometimes difficult to differentiate from lip melanoma (MEL).ObjectivesTo report in vivo reflectance confocal microscopy (RCM) features of normal lips of different phototypes and to identify features that assist in distinguishing MEL from MAC using dermoscopy and RCM.Design, setting, and participantsFor this retrospective observational study, 2 groups of patients from 2 tertiary referral centers for melanoma (Sydney Melanoma Diagnostic Centre and Melanoma Institute Australia) were recruited between June 2007 and January 2015. Group 1 included patients with normal lips and different phototypes, and Group 2 consisted of patients with MAC and MEL; RCM and dermoscopy were used for lips analysis.Main outcomes and measuresOverall, 92 RCM features were correlated with clinical history, dermoscopic images, and histopathology in all patients with MEL and 5 patients with MAC.ResultsImages from the vermillion and/or mucosal part of the lip were recorded from 10 patients with clinically normal lips (mean [SD] age, 34.5 [6.1] years), 16 patients with MAC (mean [SD] age, 49.6 [17.9] years), and 5 patients with 6 cases of MEL (1 patient had a recurrent lesion; mean [SD] age, 56.2 [15.5] years). In normal lips, the draped pattern-a previously described MAC RCM feature-was identified in all cases. In MEL, the following findings were frequent and significantly different from MAC: epidermal disarray; pagetoid infiltration of dendritic and/or round cells; a nonspecific architectural pattern at the dermoepidermal junction (DEJ); nonhomogenously distributed papillae; continuous (lentiginous) proliferation of cells with marked atypia at the DEJ, especially in interpapillary spaces; a higher number of dendritic cells (especially roundish); and atypical round cells at the DEJ. The cellular body area of dendritic cells was about the double in MEL compared with MAC. An RCM lip algorithm was developed that provided 100% sensitivity and 88% specificity for the diagnosis of MEL of the vermillion and mucosal part of the lip. With dermoscopy, MAC were correctly classified as benign in 13 of 16 cases (81%) and MEL were classified as equivocal or malignant in 5 of 6 cases (83%).Conclusions and relevanceReflectance confocal microscopy can assist in the differential diagnosis of lip MEL and MAC. An RCM Lip Score that we developed based on study results is proposed and needs to be validated on an independent data set.

Project description:Importance:The limited tissue sampling of a biopsy can lead to an incomplete assessment of basal cell carcinoma (BCC) subtypes and depth. Reflectance confocal microscopy (RCM) combined with optical coherence tomography (OCT) imaging may enable real-time, noninvasive, comprehensive three-dimensional sampling in vivo, which may improve the diagnostic accuracy and margin assessment of BCCs. Objective:To determine the accuracy of a combined RCM-OCT device for BCC detection and deep margin assessment. Design, Setting, and Participants:This pilot study was carried out on 85 lesions from 55 patients referred for physician consultation or Mohs surgery at Memorial Sloan Kettering Skin Cancer Center in Hauppauge, New York. These patients were prospectively and consecutively enrolled in the study between January 1, 2017, and December 31, 2017. Patients underwent imaging, with the combined RCM-OCT probe, for previously biopsied, histopathologically confirmed BCCs and lesions clinically or dermoscopically suggestive of BCC. Only patients with available histopathologic examination after imaging were included. Main Outcomes and Measures:Improvements in sensitivity, specificity, and diagnostic accuracy for BCC using the combined RCM-OCT probe as well as the correlation between OCT-estimated depth and histopathologically measured depth were investigated. Results:In total, 85 lesions from 55 patients (27 [49%] were female and 28 [51%] were male with a median [range] age of 59 [21-90] years) were imaged. Imaging was performed on 25 previously biopsied and histopathologically confirmed BCCs and 60 previously nonbiopsied but clinically or dermoscopically suspicious lesions. Normal skin and BCC features were correlated and validated with histopathologic examination. In previously biopsied lesions, residual tumors were detected in 12 of 25 (48%) lesions with 100% sensitivity (95% CI, 73.5%-100%) and 23.1% specificity (95% CI, 5.0%-53.8%) for combined RCM-OCT probe. In previously nonbiopsied and suspicious lesions, BCCs were diagnosed in 48 of 60 (80%) lesions with 100% sensitivity (95% CI, 92.6%-100%) and 75% specificity (95% CI, 42.8%-94.5%). Correlation was observed between depth estimated with OCT and depth measured with histopathologic examination: the coefficient of determination (R2) was 0.75 (R = 0.86; P < .001) for all lesions, 0.73 (R = 0.85; P < .001) for lesions less than 500 μm deep, and 0.65 (R = 0.43; P < .001) for lesions greater than 500 μm deep. Conclusions and Relevance:Combined RCM-OCT imaging may be prospectively used to comprehensively diagnose lesions suggestive of BCC and triage for treatment. Further validation of this device must be performed on a larger cohort.