Mobile Deep Learning Framework for Real-Time Detection of Skin Allergies and Fungal Infections Using GoogleNet

Abstract

Dermatological disorders are represented by skin allergies and fungus, and they represent a significant percentage of this group of issues across the globe; however, the equitable access to the specialty care is not evenly distributed. Early diagnosis is necessary to prevent long-term complications, but the traditional modalities of the diagnosis are lengthy and expensive. The paper presents a detailed deep-learning-based image identification system that is designed to detect allergies and early-stage fungus on the skin using the GoogleNet architecture in MATLAB. Transfer learning can be used to reduce the scarcity of medical datasets and still maintain good classification accuracy. The suggested architecture will use a mobile application elaborated by Simulink support packages and Android Studio, hence allowing real-time inference with smartphone cameras. Standardisation of input data to the convolutional neural network (CNN) involves preprocessing protocols such as image resizing, image normalisation and image augmentation. The classifier differentiates between healthy epidermis, allergic response, and fungus before providing immediate warning, which can then be used to provide immediate medical advice. This framework is based on the prioritisation of portability, scalability and computational efficiency with the aim of reducing the use of specialised clinicians to make provisional diagnoses.

Introduction

This study presents a lightweight, mobile-based deep learning system for automated dermatological diagnosis aimed at addressing the global shortage of dermatologists and the rising burden of skin diseases. Dermatological conditions, ranging from allergic reactions and fungal infections to life-threatening malignancies, require early detection to prevent severe physical and psychological consequences. However, existing AI-driven diagnostic systems face significant challenges, including limited and imbalanced medical datasets, high computational demands of deep convolutional neural networks (CNNs), hardware constraints on mobile devices, and bias across diverse skin tones.

To overcome these limitations, the proposed system employs a transfer-learning framework based on GoogLeNet, a computationally efficient CNN architecture featuring Inception modules for multi-scale feature extraction. The model is pretrained on ImageNet and fine-tuned for classifying normal skin, allergic reactions, and fungal infections. Data augmentation techniques (rotation, zooming, flipping) are applied to mitigate overfitting and class imbalance. Training is conducted in MATLAB using optimized hyperparameters and dataset partitioning (70% training, 15% validation, 15% testing).

The system architecture follows a pipeline structure: image acquisition via smartphone camera, preprocessing (resizing, normalization, noise reduction), feature extraction using GoogLeNet, SoftMax-based classification, and real-time result display. Deployment is achieved through MATLAB and Simulink integration with Android devices, enabling offline, real-time inference without reliance on cloud connectivity.

Performance evaluation uses metrics such as accuracy, precision, recall (sensitivity), specificity, and F1-score. Literature-based benchmarks suggest achievable accuracy around 89% under adequate training conditions. The system demonstrates potential as a cost-effective preliminary screening tool, especially in low-resource environments.

Despite its advantages, limitations include sensitivity to lighting and image quality, difficulty detecting low-contrast lesions, and potential bias due to underrepresentation of diverse skin tones in training datasets. The study concludes that mobile-friendly, transfer-learning-based AI systems offer a scalable and accessible approach to early dermatological screening while highlighting the need for fairness-conscious validation and dataset expansion.

If needed, I can also provide:

• A short journal-style abstract (200 words)

• A conference presentation summary

• Or a simplified version for non-technical audiences

Conclusion

In the present research, a deep learning model has been identified to detect skin allergies and fungal infections in real-time using the GoogleNet architecture. With the capabilities of transfer learning in a MATLAB-Simulink setup, we developed a feasible procedure of implementing state-of-the-art convolutional neural networks in Android portable devices. The system architecture explicitly covers the need in portable, quick, and affordable dermatological screening. Though GoogleNet provides a strong balance between the ability to extract features and the computational efficiency, continuing issues with the balance of datasets and skin-tones represent urgent priorities in the further development. As a result, the presented system can be viewed as the step forward in AI-controlled healthcare, and it can empower people with the ability to detect issues early on and promote the overall aim of medical diagnostics accessibility. The combination of strict preprocessing process, effective model structure, and convenient mobile implementation provides a strong base of future innovations in teledermatology Continued studies are needed to take a step forward and include structured clinical information, including family history, as suggested by Jeong et al. to increase decision-making fidelity and diagnostic reliability. Furthermore, the later versions can involve hardware acceleration methods to reduce the amount of energy used on the mobile systems.

References

[1] H. K. Jeong, C. Park, R. Henao, and M. Kheterpal, “Deep Learning in Dermatology: A systematic review of current approaches, outcomes, and limitations,” JID Innovations, vol.3, no.1, p. 100150, Aug. 2022, doi: 10.1016/j.xjidi.2022.100150. Available: https://doi.org/10.1016/j.xjidi.2022.100150 [2] Yao, Peng, Shen, Shuwei, Xu, Mengjuan, Liu, Peng, Zhang, Fan, Xing, Jinyu, Shao, Pengfei, Kaffenberger, Benjamin, Xu, Ronald X., \"Single Model Deep Learning on Imbalanced Small Datasets for Skin Lesion Classification,\" IEEE TRANSACTIONS ON MEDICAL IMAGING, 2021, 2021. doi:10.1109/TMI.2021.3136682 [3] Gokhale, Vinayak, Zaidy, Aliasger, Chang, Andre Xian Ming, Culurciello, Eugenio, \"Snowflake: A Model Agnostic Accelerator for Deep Convolutional Neural Networks,\" 2017. https://arxiv.org/pdf/1708.02579v1 [4] R. Daneshjou et al., “Disparities in dermatology AI performance on a diverse, curated clinical image set,” Science Advances, vol. 8, no. 32, p. eabq6147, Aug. 2022, doi: 10.1126/sciadv.abq6147. Available: https://doi.org/10.1126/sciadv.abq6147 [5] Chiu, Ming-Chang, Wang, Yingfei, Kuo, Yen-Ju, Chen, Pin-Yu, \"DDI-CoCo: A Dataset For Understanding The Effect Of Color Contrast In Machine-Assisted Skin Disease Detection,\" 2024. https://arxiv.org/pdf/2401.13280v1 [6] A. R. F. Santos, K. R. T. Aires, and R. M. S. Veras, “Aspects of Lighting and Color in Classifying Malignant Skin Cancer with Deep Learning,” Applied Sciences, vol. 14, no. 8, p. 3297, Apr. 2024, doi: 10.3390/app14083297. Available: https://doi.org/10.3390/app14083297 [7] Banerjee, Arnab, Das, Nibaran, Nasipuri, Mita, \"Skin Diseases Detection using LBP and WLD- An Ensembling Approach,\"2020. https://arxiv.org/pdf/2004.04122v1.

Copyright

Copyright © 2026 Vinish S, Vijayakumar J, Mervin Paul Raj M. 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.