SkinAI-Skin Disease Detection and Classification Using Machine Learning by IRJET Journal

SkinAI-Skin Disease Detection and Classification Using Machine Learning

Chandrakant

Barde1 , Priyanka Bhamare2 , Krushna Shinde3 , Prachi Pawar4 , Sakshi Chaudhari5

1 AssistantProfessor,Dept.ofComputerEngineering, GES’s R.H.SapatCollegeofEngineering,Nashik, Maharashtra,India

2,3,4,5 Student,Dept.ofComputerEngineering, GES’s R.H.SapatCollegeofEngineering,Nashik,Maharashtra,India

Abstract - TheSkinAIresearchexploresacutting-edge approachtodetectingandclassifyingskindiseasesusing aConvolutionalNeuralNetwork(CNN)integratedintoa webbased platform. Skin diseases are widespread, but early detection, especially of critical conditions like melanoma, is essential for effective treatment. Traditionaldiagnosticmethodsareofteninaccessibleto individuals in remote areas, creating a demand for automated solutions. Our system allows patients and healthcareprofessionalstouploadorcaptureimagesof skinlesionsviathewebandreceivereal-timediagnostic resultsandsuggestedtreatments.Thebackendutilizesa CNN model trained on dermatological datasets to classifycommonskindiseaseslikeeczema,psoriasis,and melanoma.Thispaperprovidesadetailedbreakdownof the algorithm, system architecture, and model, and comparesourapproachwithexistingmachinelearning methods. Results demonstrate high accuracyindisease detection anda user-friendly experience, makingthis a promisingtoolfortelemedicineapplications.

Key Words: Skin Disease Detection, Symptom-Based Recommendation, Machine Learning, Home Remedies Recommendation

1.INTRODUCTION

Inrecentyears,skindiseaseshavebecomeoneofthe most prevalent health concerns worldwide, affecting millionsofindividualsacrossvariousagegroupsand demographics. With conditions ranging from benign issues like acne to more severe cases such as melanoma, the demand for accurate, efficient, and timely diagnosis has never been greater. Dermatologicalexaminationsarecriticalfordiagnosing theseconditions,butaccesstospecializedcarecanbe limited, particularlyinremoteor underserved areas. This limitation emphasizes the need for automated solutionsthatcanassistbothpatientsandhealthcare professionalsintheearlydetectionandclassificationof skindiseases.

ArtificialIntelligence(AI),andmorespecifically,deep learning, has revolutionized many fields, including healthcare.ConvolutionalNeuralNetworks(CNNs),a typeofdeeplearningmodelparticularlyadeptatimage recognitiontasks,haveshowngreatpromiseinmedical diagnostics, especially in dermatology. Skin disease diagnosisreliesheavilyonvisualassessments,making it an ideal candidate for automation through AI. By leveraging CNNs, it becomes possible to classify skin diseases based on images, allowing for a faster and moreaccuratediagnosisprocess.Thisresearchfocuses on developing a web-based system for skin disease detection and classification, integrating CNNs to automate the diagnostic process. Users, including patients and medical professionals, can upload or captureimagesofskinlesions,whicharethenanalyzed inreal-timebyaCNNmodeltrainedondermatological datasets. The system not only identifies the skin condition but also suggests potential treatments, offeringapreliminarydiagnosisthatcanguidefurther medical consultations. The primary objective of this research is to create an accessible, user-friendly platformthatcanbeutilizedbothasadiagnostictool andaneducationalresource.Byautomatingtheinitial assessmentofskindiseases,thesystemaimstobridge the gap between patients and dermatologists, particularly in areas where healthcare services are scarce.Thispaperoutlinesthedevelopmentprocessof the system, the CNN model architecture, and the technologicalframeworksused,whilealsopresentinga detailed comparison with existing methods for skin diseasedetection.

1.1 Features of the Skin disease detection and classification web application: Oursystemisdesignedtoautomatethedetection andclassificationofskindiseases,withafocuson accessibilityandeaseofuse.Keyfeaturesinclude: User Roles: 1.

Patients:Canuploadorcaptureimagesbuthave restrictedaccesstodatamodification.

Volume:12Issue:01|Jan2025 www.irjet.net

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056 p-ISSN:2395-0072

2.Doctors:Can upload images, modify data, and accessdetailedanalytics.

3. Image Upload and Capture: Users can either upload images from their device's gallery or captureliveimagesthroughawebcam.

4. Disease Detection Using CNNs: After image submission, the system preprocesses the image andrunsitthroughaCNNmodelthatclassifiesthe diseaseandprovidestreatmentsuggestions.

5. Real-Time Processing: The system provides immediate feedback on the uploaded image, displaying the predicted disease and recommendedtreatment.

6.HistoryTracking:Userscanaccessahistoryof their last four diagnoses, enabling continuous monitoringofskinconditions.

7.Data Security: The system ensures privacy by storinguserdatasecurelyinarole-baseddatabase system,usingencryptiontechniques.

1.2 Advantages of a Skin disease detection:

Theplatformisavailableonline,providingusersin remoteorunderservedareaswithaccesstobasicskin diseasediagnosiswithoutneedingadermatologist.

1.Efficiency:Byautomatingthediagnosticprocess, thesystemdramaticallyreducesthetimerequiredfor patientstoreceivepreliminarydiagnoses.

2.Cost-Effective:Theplatformeliminatesthe need forexpensiveequipmentorrepeatedin-personvisits to dermatologists, offering a low-cost alternative for initialassessments.

3. Scalability: The modular design of the system allows for the addition of new diseases, treatment protocols,anddiagnosticfeatureswithoutsignificant overhauls.

4.AutomatedDataHandling:Historicaldatarelated touserimagesanddiagnosesareautomaticallystored andaccessible,eliminatingtheneedformanualrecordkeeping.

5.Role-BasedAccessControl:Doctorsandpatients havedifferentaccesslevels,ensuringthatdatacanbe updatedonlybyauthorizedusers.

2. LITERATURE REVIEW

The use of machine learning and deep learning techniques in skin disease detection has garnered significant attention in recent years, with several studiesshowcasingtheadvantagesandchallengesof variousapproaches.

OnesuchstudybyC.Nallusamy(2023),publishedin the Journal of Population Therapeutics & Clinical Pharmacology,focusesondeeplearningmodelsforthe detection of melanoma, one of the deadliest types of skin cancer. The research demonstrates that deep learningalgorithms,particularlyconvolutionalneural networks(CNNs),cansignificantlyimprovethespeed andaccuracyofmelanomadetectionwhencompared to traditional diagnostic methods. Nallusamy's study emphasizes the importance of early detection and suggeststhatdeeplearningcouldbeaviablemethodto integrateintoclinicalpractices,reducinghumanerror anddiagnosticdelays.Learningtechniques,especially in terms of classification accuracy and feature extraction capabilities. The study also points out the limitations of machine learning models, which often require manual feature selection, making them less adaptable to new data. In contrast, deep learning models,particularlyCNNs,canautomaticallylearnand extract important features from images, leading to higher accuracy, especially for complex skin disease images.

In another study of, Md. Al Mamun and Mohammad Shorif Uddin (2022) proposed a machine learningbased solution for skin disease detection and classificationusingimagesegmentation.Publishedin Elsevier, their work discusses how segmentation techniques can enhance the accuracy of skin disease detection by isolating the region of interest (the affected skin area) before feeding the image into a classificationmodel.Theirapproachhelpedtoreduce thenoiseinimagesandimprovethemodel'sabilityto focus on relevant parts of the skin lesion, leading to more accurate diagnoses. However, their study also recognizedthelimitationsofmachinelearningmodels when compared to deep learning, as the latter offers greaterflexibilityandpowerindealingwithlargeand diversedatasets.

Another significant contribution is from Suganya R. (2016),whoexploredtheapplicationofanAutomated Computer-Aided Diagnosis (CAD) system for skin lesionsindermoscopyimages. Presented at the Fifth International Conference on Recent Trends in Information Technology, the research outlines how CAD systems, combined with neural networks, can improve the precision of diagnosing skin conditions, particularlymelanoma.Suganya’sworkhighlightsthat automated systems can significantly reduce the time andeffortrequiredbydermatologistsforpreliminary diagnoses,makingthemavaluabletoolintelemedicine

Volume:12Issue:01|Jan2025 www.irjet.net

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056 p-ISSN:2395-0072

andlarge-scalehealthcaresystems.Herresearchalso discusses the benefits of integrating deep learning models, which can generalize better across different skintypesanddiseasevariations.

Lastly,MarínC.,AlférezG.H.,andGonzálezV.(2015) conducted a study on the non-invasive detection of melanoma using artificial neural networks (ANNs). PublishedbySpringer,theirresearchpresentsanovel method of distinguishingbetweenmelanoma lesions andbenignnevithroughimagerecognitiontechniques. Theauthorsarguethatusingneuralnetworkscanhelp automate the detection process, providing quickand reliableresultsthatcouldaiddermatologistsinclinical settings.Whiletheirresearchfocusesprimarilyonthe ANN approach, it opened up avenues for exploring more advanced neural network models like CNNs, which are now more commonly used due to their superiorperformanceinimageclassificationtasks.

Together,thesestudiesprovidestrongevidenceforthe effectiveness of deep learning models, particularly CNNs, in the field of skin disease detection. While machine learning models like decision trees and support vector machines have shown promise, their performanceisoftenoutpacedbyCNNs,whichexcelin processingcompleximagedata.Thegrowingbodyof literaturesuggeststhatdeeplearningcanplayapivotal role in advancing telemedicine applications and providing faster, more accurate diagnostic tools for dermatologistsandpatients.

3. PROPOSEDSYSTEM

Fig-1:ProposedSystemArchitecture

Theskindiseasedetectionsystemincorporates severaltechnologiesthatworktogetherseamlessly toprovideanintuitiveuserexperienceandaccurate diseaseclassification.

3.1 WORKFLOW FOR IMAGE PROCESSING MODULE:

Step1:Useruploadsorcapturesanimageviatheweb interface.

Step 2: The image is preprocessed (resized, normalized)toaformatacceptablefortheCNNmodel (e.g.,224x224pixels).

Step3:Thepreprocessedimageissenttothetrained CNNmodel, which performsdiseaseclassificationby extractingkeyfeaturesthroughseveralconvolutional layers.

Step4:Thebackendreceivesthepredicteddiseaseand medicationsuggestions.

Step5:Theresultsaredisplayedonthefrontend,and the user is also given an option to review previous diagnosesthroughahistoryfeature.

3.2 WORKFLOW FOR DISEASE DETECTION MODULE :

Step1:UserInteraction:

1.Login/Signup:Userselectswhethertologinorsign up. ○ Role selection (Patient/Doctor) during signup determinesaccesspermissions.

2. Image Upload or Capture: The user can upload or captureanimageviatheinterface.

Step2:ImagePreprocessing:

1. The uploaded image is received by the backend throughanHTTPPOSTrequest.

2. Image is stored temporarily and preprocessed (resized, normalized). 3. The preprocessed image is transformedintonumpyarray.

Step3:CNNModel:

ThepreprocessedimageispassedtotheCNNmodel. 1.ModelPrediction:TheCNNmodelpredictstheskin diseasebasedonthefeaturesextractedfromtheimage.

2. Suggested Treatment: Along with the disease prediction,thesystemsuggestspossibletreatments.

Step4:BackendResponse:

1. Thepredicteddiseaseandtreatmentsuggestions arereturnedasaJSONobject.

2.Theresponseissentbacktothefrontendfordisplay totheuser.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056 p-ISSN:2395-0072

Volume:12Issue:01|Jan2025 www.irjet.net

4. METHODOLOGY

The methodology involves collecting a diverse skin diseaseimagedataset,preprocessingitwithresizing, normalization, and augmentation, and extracting features using traditional methods or deep learning (CNNs like ResNet, VGG). Machine learning models (SVM, Random Forest) or deep learning models classifydiseases,withtrainingoptimizedusingcrossvalidation and performance evaluated via accuracy, precision, recall, and AUC-ROC. Model interpretability is enhanced using Grad-CAM and SHAP.Ifdeployed,themodelisintegratedintoaweb ormobileapp.Ethicalconsiderations,includingdata privacy and fairness, are addressed for responsible AIuse.

4.1 Data Collection

 Obtainapubliclyavailabledatasetsuchas ISIC (International Skin Imaging Collaboration),DermNet,HAM10000, ora hospital/private dataset (with ethical approval).

 Datashouldincludeimagesofvariousskin diseaseswithcorrespondinglabels.

 Ensuredatasetdiversityintermsofskin tones, age groups, and environmental conditions.

4.2. Data Preprocessing

 Image Resizing: Resize images to a fixed dimension(e.g.,224x224fordeeplearning models).

 Normalization:Normalizepixelvalues(e.g., between0and1or-1and1).

 Data Augmentation: Apply transformations likerotation,flipping,brightnessadjustment, and noise addition to improve model generalization.

 Class Balancing: Use techniques such as oversampling, under sampling, or Synthetic MinorityOver-samplingTechnique(SMOTE) ifdatasetclassesareimbalanced.

 Segmentation(ifrequired):Usesegmentation techniques like U-Net to focus on the lesion area.

4.3. Feature Extraction

 TraditionalMachineLearningApproach: Use handcrafted features such as color histograms,texturedescriptors(GLCM,LBP), andshapefeatures.

 Deep Learning Approach: Use pre-trained CNNmodelslikeResNet,VGG,EfficientNet, MobileNetforautomaticfeatureextraction.

4.4. Model Selection

 TraditionalMachineLearningModels: Train classifiers like Support Vector Machine (SVM), Random Forest (RF), k-Nearest Neighbors (k-NN), Decision Trees, or XGBoostonextractedfeatures.

 Deep Learning Models: Implement deep learning architectures such as CNN (Convolutional Neural Networks), or use transferlearningwithpre-trainedmodels.

 Hybrid Approaches: Combine deep learning withtraditionalclassifiers(e.g.,usingCNNfor featureextractionandSVMforclassification).

4.5. Explainability & Model Interpretation

 Grad-CAM (Gradient-weighted Class ActivationMapping):Visualizewhichpartsof theimageinfluencedthemodel’sdecision.

 SHAP(ShapleyAdditiveExplanations)or LIME(LocalInterpretableModel-agnostic Explanations):Explainmodelpredictions.

4.6. Deployment

 Convert the model into a web-based applicationormobileapplicationusingFlask, FastAPI, or TensorFlow Lite for real-world usability.

 Deploythemodelusingcloudplatformslike GoogleCloud,AWS,orMicrosoftAzure.

4.7. Ethical Considerations

 Ensuredataprivacyandsecurity.

 Addressbiasinskindiseasedetection(e.g., ensuring fair performance across different skintones).

 FollowmedicalAIregulationsandguidelines.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056 p-ISSN:2395-0072

Volume:12Issue:01|Jan2025 www.irjet.net

5. EXPERIMENTAL RESULT

areastoreceiveanearlydiagnosisbutalsoservesasa valuabletoolformedicalpractitionersseekingaquick referenceorpre-screeningtoolforpatients.Although the results have been promising, there is room for improvement. Expanding the dataset to include a broaderrangeofskinconditionsandenhancingthe system’sabilitytohandleimageswithvaryingquality levelswillimproveaccuracy.Furthermore,futurework should focus on refining the system's predictive capabilities by incorporating user feedback and retraining the CNN model with new data. Enhancing securityfeaturestocomplywithhealthcareregulations such as GDPR and HIPAA will also be crucial as the system scales for wider use. Overall, this research marks a significant step forward in using artificial intelligence for healthcare solutions, particularly in dermatology. The integration of CNNs for automated skindiseasedetectioncanrevolutionizetelemedicine and provide a cost-effective alternative for patients seekingearlydiagnosis.

7.REFERENCES

CONCLUSIONS

The research on automating skin disease detection using Convolutional Neural Networks (CNNs) demonstrates the immense potential of AI-driven technologiesinrevolutionizinghealthcarediagnostics. By developing a web-based platform that integrates CNNs, we have provided a practical solution for detectingandclassifyingcommonskindiseasessuchas eczema, psoriasis, and melanoma. This system addresses critical challenges in dermatological care, particularlythelackofimmediateaccesstospecialists, by offering users patients and healthcare professionals alike a reliable, efficient, and easily accessibletoolforpreliminarydiagnosis.Thesuccess of this project lies in its ability to process images throughCNN-basedimagerecognitionmodels,offering real time feedback on disease classification and treatment recommendations. Additionally, the inclusionofautomatedschedulingfeaturesimproves the system's usability, ensuring continuous health monitoringandtimelyinterventions.Thissystemnot

The proposed CNN model for skin disease detection was trained and evaluated using a dataset of dermatological images. The model demonstrated strong performance in classifying various skin conditions with high accuracy. Standard evaluation metrics,includingaccuracy,precision,recall,andF1score, were used to assess its effectiveness. The confusion matrix analysis indicated that the model performed well in detecting common skin diseases but faced occasional misclassifications in visually similar conditions. In comparison with existing models, our approach outperformed traditional machine learning techniques and showed competitive performance against advanced deep learning architectures. However, certain challenges were observed, particularly in differentiating between diseases with overlapping visual features, such as eczema and psoriasis. Additionally, class imbalance in the dataset influenced the model's predictions, suggesting the need for further improvements through data augmentation and ensemble learning techniques. Future work will focusonenhancingthedatasetdiversity,refiningthe model architecture, and incorporating explainability methods such as Grad-CAM to improve interpretabilityandtrustinthemodel’spredictions. theutilityofautomatedsystemsindermatology.

[1]Nallusamy,C.(2023)."SkinDiseaseDetectionBased onDeepLearning."JournalofPopulationTherapeutics &ClinicalPharmacology.Thisstudyhighlightstheuse ofdeeplearningmodels,particularlyCNNs,intheearly detection of melanoma, significantly improving diagnosticspeedandaccuracy.

[2] Bose, P., & Bandyopadhyay, S. K. (2021). "Skin Disease Detection: Machine Learning vs Deep Learning." Preprints. This paper provides a comparative analysis of traditional machine learning and deep learning approaches, emphasizing the superioraccuracyandperformanceofCNNsinimagebaseddiseaseclassification.

[3]Ahammed, M., &Uddin,M. S. (2022). "AMachine Learning Approach for Skin Disease Detection and ClassificationUsingImageSegmentation."Elsevier.The research proposes an image segmentation-based methodtoenhancetheaccuracyofmachinelearning modelsforskindiseaseclassification.

[4] Suganya, R. (2016). "An Automated ComputerAided Diagnosis of Skin Lesions Detection and Classification for Dermoscopy Images." Fifth International Conference on Recent Trends in Information Technology. This study presents a CAD systemthatusesimageanalysistechniquestoimprove theprecisionofskinlesiondiagnosis,demonstrating onlyempowersindividualsinremoteandunderserved

Volume:12Issue:01|Jan2025 www.irjet.net

[5]Marín,C.,Alférez,G.H.,Córdova,J.,&González,V. (2015). "Detection of Melanoma Through Image RecognitionandArtificialNeuralNetworks."Springer. The research discusses a non-invasive approach to

melanoma detection using artificial neural networks, emphasizingtheroleofimagerecognitiontechnologies inimprovingdiagnosticoutcomes.

[6] Tiwari, S., Sharma, A., & Rani, S. (2021). "Smart BuildingManagementSystemusingIoT."International JournalofEngineeringResearch&Technology(IJERT). Thoughfocusedonbuildingmanagementsystems,this studyservesasanexampleofusingIoTandAIforrealtimemonitoringandautomation,illustratingparallels inhowAIcanoptimizeworkflowsinotherfieldslike healthcare.

[7]Sharma,S.etal.(2022)."AdvancedTechniquesin Dermatology Using Machine Learning." IEEE International Conference on Machine Learning for Healthcare.Thispaperreviewstheimpactofmachine learningmodelsonimprovingdiagnosticprocessesin dermatology, supporting the importance of AI integrationinmedicalpractice.

[8] Ghahramani, A., et al. (2018). "Skin Disease ClassificationUsingDeepLearningandArtificialNeural Networks." Journal of Biomedical Imaging and Bioengineering. This study explores the accuracy of deep learning methods in classifying complex skin diseasesandemphasizesthevalueofneuralnetworks inimprovingdiagnosticaccuracy.