Clinico-pathological Analysis Of Eyelid Tumors And New Diagnosis Approaches Based On Artificial Intelligence

Purpose: To describe the clinico-pathological characteristics of patients with common eyelid tumors and tumor-like lesions.Methods: A total of 5146 cases of eyelid lesions in the Second Affiliated Hospital of Zhejiang University(ZJU-2)were reviewed from 2000 to 2018,being classified by histogenesis and pathologic diagnosis.Age-specific and gender-specific incidence constitutions,time trends and distribution in different age groups were calculated.We further compared data from other countries containing different races.Results: Histopathological results of the excised lesions are presented in Table1.Benign eyelid tumors accounted for 85.08% of all cases while malignant eyelid tumors accounted for 14.92%.Among the benign tumors,nevus was most common(33.07%),followed by squamous papilloma(12.31%),and basal cell papilloma(9.55%).The R2 value of linear regression in patient number of benign lesions were 0.946(P<0.01)for male and 0.914(P<0.01)for female.More than 33.60%(1471/4378)were made up by patients younger than 40 years.The number of patients undergoing removal of benign lesions decreased with age.Among the malignant lesions,basal cell carcinoma(BCC)was most prevalent(371,48.70%),followed by sebaceous gland carcinoma(263,34.24%),squamous cell carcinoma(95,12.37%)and large majority(81.8%)occurred in patients above 60 years.Conclusions: Over the past nineteen years,most eyelid tumors occurred at ZJU-2 were benign lesions.The number of patients presenting with benign lesions increased in both genders,especially among young females who were more likely to request surgeries.Among malignant lesions,BCC remains the most common type but appears a much lower incidence comparing with light-skinned population.The spectrum of tumors was determined by race,socioeconomic situation and patients’ willingness to surgery.Background/Aims: To develop a deep learning system(DLS)that can automatically detect malignant melanoma in the eyelid from histopathological sections with colossal information density.Methods:Setting: Double institutional study.Study Population: We retrospectively reviewed 225,230 pathological patches(small section cut from pathologist-labelled area from a hematoxylin-eosin [H&E] image),cut from 155 H&E stained whole-slide images(WSIs).Observation Procedures: Labelled gigapixel pathologic WSIs were used to train and test a model designed to assign patch-level classification.Using malignant probability from a convolutional neural network(CNN),the patches were embedded back into each WSI to generate a visualization heatmap and leveraged a Random Forest model to establish a WSI-level diagnosis.Main Outcome Measure(s): For classification,the area under the receiver operating characteristic curve(AUC),accuracy,sensitivity,and specificity were used to evaluate the efficacy of the DLS in detecting malignant melanoma(MM).Results: For patch diagnosis,the model achieved an AUC of 0.989(95% confidence interval [CI]: 0.989–0.991),with an accuracy,sensitivity,and specificity of 94.9%,94.7%,and 95.3%,respectively.We displayed the lesion area on the WSIs as graded by malignant potential.For WSI,the obtained sensitivity,specificity,and accuracy were 100%,96.5%,and 98.2%,respectively,with an AUC of 0.998(95% CI: 0.994–1.000).Conclusion: Our DLS,which uses artificial intelligence,can automatically detect MM in histopathological slides and highlight the lesion area on WSIs using a probabilistic heatmap.In addition,our approach has the potential to be applied to the histopathological sections of other tumor types.Purpose: To build up the proteomic database of common eyelid tumors based on PCT-SWATH,analyze differential proteins and develop a computer aided diagnosis system using Artificial Intelligence for potential primary screening.Methods: Retrospectively included 332 formalin fixed paraffin embedded(FFPE)samples from 216 patients(233 training cohort/99 testing cohort),including benign pigmented nevus,squamous cell papilloma,basal cell papilloma,basal cell carcinoma,squamous cell carcinoma,sebaceous gland carcinoma,malignant melanoma and normal eyelid tissue(from cosmetic patients).Take small punches(diameter<0.8 mm)from FFPE.A highly reproducible Pressure Cycling Technology coupled with SWATH(PCT-SWATH)proteomics workflow was applied on the eyelid tumor samples after preprocessing.After quality control,we used genetic algorithm to select best proteomic feature combination and then added features with biological significances based on stochastic subspace identification.Finally,utilizing multilayer perceptron to achieve the whole construction of this diagnostic classification system.Results: Total 60360 protein_groups and 4636 proteins were detected in this work.Normal tissues had a significant lower protein abundance in 3482 proteins(P <0.05),and higher expression in 133 proteins(P <0.05).The DLS established in this study recruited 18 features from total 4636 proteins for final classification.For the binary classification(Normal vs Tumor),the algorithm achieved 0.973 area under the curve,0.949 accuracy(same for balanced_accuracy),86.4% sensitivity and 97.4% specificity.In 7-classification of individual disease,our model achieved 0.828 overall accuracy(0.818 balanced_accuracy).Conclusion: Our DLS,which uses PCT-SWATH technique,established proteomic landscape of common eyelid tumors.Among which,the tumor tissues had a significant higher protein abundance of normal eyelid tissues.Using novel Artificial Intelligence algorithm,we built a DLS which could diagnose common eyelid tumors with high accuracy based on proteomic data,independent of traditional pathological features.The differentiated expressed proteins among different types and 18 features selected by DLS could be implemented for future mechanism investigation and as potential biomarker for the development of targeted drugs.