Skin Cancer Detection Using Deep Learning

Abstract

The prop.osed project, Skin Cancer Detection Usi.ng Deep Lear.ning, focu.ses on desig.ning and implem.enting an automated sys.tem capa.ble of identi.fying and classifying skin lesi.ons as ben.ign or malig.nant usi.ng image proce.ssing and neu.ral networ.k-based classification techn.iques. The main objec.tive of this proj.ect is to aid dermato.logists and medical practit.ioners by provi.ding a fast, reli.able, and accu.rate tool for ear.ly detec.tion of skin can.cer, espec.ially mela.noma. The system empl.oys Convolu.tional Neu.ral Netw.orks (CNNs) and tran.sfer lear.ning mod.els such as VGG.16, ResN.et50, and Mobile.NetV2 for ima.ge feature extra.ction and classif.ication. The data.set used is the HAM1.0000 skin lesion dataset, conta.ining thous.ands of dermato.scopic ima.ges of various skin les.ion typ.es. The proposed sys.tem prepro.cesses ima.ges thro.ugh normali.zation, augmen.tation, and segmen.tation, ensu.ring bet.ter general.ization and robustness. Experi.mental resu.lts demons.trate high accu.racy, preci.sion, and rec.all, prov.ing the effecti.veness of deep learnin.g-based mod.els for medi.cal ima.ge anal.ysis. This work highl.ights how AI-d.riven diagn.ostic too.ls can red.uce hum.an err.or and impr.ove ear.ly can.cer detection rat.es, ultim.ately sav.ing liv.es.

Introduction

Skin cancer is one of the most common and dangerous cancers worldwide, and early detection is essential for improving survival rates. Traditional diagnosis relies on visual examination and biopsy, which can be slow and prone to human error. With advancements in Artificial Intelligence and Computer Vision, automated image-based detection has become a promising approach.

The project focuses on developing a CNN-based deep learning model to classify skin lesion images as benign or malignant, using the HAM10000 dataset. The system includes preprocessing steps such as resizing, normalization, augmentation, and segmentation to enhance image quality. Transfer learning models like VGG16, ResNet50, and MobileNetV2 are used for efficient feature extraction. The model is trained using binary cross-entropy loss and evaluated with accuracy, precision, recall, and F1-score metrics.

A literature survey highlights major contributions in AI-driven dermatology, including dermatologist-level performance by CNNs (Esteva et al., 2017) and benchmark datasets like ISIC and HAM10000. Recent works focus on hybrid architectures and lightweight, explainable AI for mobile deployment.

The project also includes a user-friendly web interface with modules such as Login, Registration, Upload & Analyze, Result Display, and Report Generation. Users can upload skin lesion images, view predictions with confidence scores, and download a structured medical report including diagnosis and hospital recommendations.

The proposed model achieved 85.4% accuracy, showing robustness to lighting variations and improved performance due to transfer learning. Overall, the system supports early identification of skin cancer and can assist clinicians as a diagnostic support tool. Future improvements may include larger datasets and more advanced deep learning architectures.

Conclusion

The prop.osed system succes.sfully dete.cts skin can.cer usi.ng deep lear.ning techn.iques. The CNN-based mod.el trai.ned on the HAM1.0000 data.set achi.eved high accu.racy in classifying skin lesi.ons as ben.ign or malig.nant. Experi.mental results demonstrated that tran.sfer lear.ning archite.ctures like ResN.et50 signifi.cantly enha.nce perfor.mance comp.ared to traditional meth.ods. FutureScope: Furt.her improv.ements can incl.ude: • Integr.ation of Explai.nable AI (XAI) to just.ify predic.tions. • Deplo.yment as a mob.ile or web-.based diagn.ostic tool for real.-time detec.tion. • Use of 3D dermato.scopic data for enha.nced diagn.ostic accu.racy. This rese.arch contri.butes tow.ard accessible, affor.dable, and AI-d.riven healt.hcare solut.ions for ear.ly skin can.cer detec.tion.

References

[1] A. Est.eva, B. Kup.rel, R. A. Nov.oa et al., “Dermatologist-level classif.ication of skin cancer with deep neural netw.orks,” Nat.ure, vol. 542, no. 7639, pp. 115.–118, 2017. [2] N. C. Code.lla, D. Gut.man, M. E. Cel.ebi et al., “Skin les.ion anal.ysis tow.ard mela.noma detec.tion: A chall.enge at the 2018 ISIC,” IEEE Transa.ctions on Medi.cal Imag.ing, vol. 38, no. 8, pp. 1685–1697, 2019. [3] P. Tsch.andl, C. Rosen.dahl, and H. Kitt.ler, “The HAM1.0000 data.set: A large colle.ction of multi-.source dermatoscopic ima.ges of com.mon pigme.nted skin lesions,” Scien.tific Data, vol. 5, pp. 1–9, 2018. [4] T. J. Brin.ker, A. Hek.ler, F. Enk et al., “Deep learning outper.formed 136 of 157 dermato.logists in a head-to-head dermoscopic mela.noma ima.ge classif.ication task,” European Jour.nal of Can.cer, vol. 113, pp. 47–54, 2019. [5] S. S. Han, I. Kim, and Y. Lim, “Hy.brid deep neu.ral netw.orks for skin les.ion classif.ication,” IEEE Acc.ess, vol. 8, pp. 101,85.8–101,869, 2020. [6] R. C. Gess.ert, T. Sent.ker, R. Made.sta et al., “Skin lesion classif.ication usi.ng ensem.bles of deep neural networks,” Medi.cal Ima.ge Anal.ysis, vol. 66, p. 101.–852, 2020. [7] M. Ahs.an, M. Alam, and M. Rahman, “Efficient skin can.cer detec.tion usi.ng CNN-based transfer lear.ning,” Proc.edia Comp.uter Scie.nce, vol. 167, pp. 2410.–2419, 2020. [8] S. Al-M.asni, M. Al-A.ntari, M. Choi et al., “Skin les.ion segmen.tation and classif.ication with deep lear.ning sys.tem,” Computer Meth.ods and Prog.rams in Biomed.icine, vol. 162, pp. 221.–231, 2018. [9] Kag.gle, “HAM.10000: Hum.an Agai.nst Mach.ine with 100.00 trai.ning images,” 2023. [On.line]. Avail.able: https://www.kaggle.com/datasets/kmader/skin-cancer-mnist-ham10000 [10] Tenso.rFlow Documen.tation, “Transfer lear.ning for ima.ge classif.ication with Tenso.rFlow and Keras,” 2024. [11] Est.eva, A., et al. (2017). Dermatologist-level classif.ication of skin can.cer with deep neu.ral netw.orks. Nat.ure, 542, 115.–118. [12] Brin.ker, T. J., et al. (2019). Compa.ring artif.icial intell.igence algor.ithms for skin can.cer diagn.osis. Front.iers in Medi.cine. [13] Haen.ssle, H. A., et al. (2018). Man vs. Mach.ine: Compa.rison of Diagnostic Perfor.mance for Mela.noma. Ann.als of Onco.logy. [14] Phil.lips, M., et al. (2019). Asses.sment of the accu.racy of deep lear.ning for detec.ting dermatological dise.ases. JAMA Dermat.ology. [15] Fer.ris, L. K., et al. (2015). Smart.phone applic.ations for skin can.cer diagn.osis: Evalu.ation of perfor.mance. JAMA Dermat.ology. [16] Lam, C., et al. (2018). Autom.ated mela.noma detec.tion usi.ng comp.uter vision. Jour.nal of Medi.cal Imag.ing.

Copyright

Copyright © 2025 Mr. Nikesh Bode, Anushka V. Korde, Deepak S. Dhale, Hrutik S. Hiwase, Kartik V. Gujarkar, Tanmay Chaudhari. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.