Deep Learning Based Disease Detection Using Facial Features

Deep Learning Based Disease Detection Using Facial Features

1st Prof.Ravindra Patil                                             2nd A S Prajwal                                           3rd Abhishek V Honnedibba

Department of Computer Science                       Department of Computer Science                Department of Computer Science

KLS Vishwanathrao Deshpande                           KLS Vishwanathrao Deshpande                      KLS Vishwanathrao Deshpande

Institute of Technology                                          Institute of Technology                                   Institute of Technology

Haliyal, India                                                                Haliyal, India                                                         Haliyal, India

rtp@klsvdit.edu.in                                                asprajwal120@gmail.com                            abhishekhonnedibba@gmail.com

4th Hemavati H Bhovivaddar                                 5th Sanjay G Hegde

Department of Computer Science                Department of Computer Science

KLS Vishwanathrao Deshpande                     KLS Vishwanathrao Deshpande

Institute of Technology                                 Institute of Technology

Haliyal, India                                                            Haliyal, India

87hemavatibhovidvaddar@gmail.com           sanjayhegde1636@gmail.com

­­­­­­ Abstract

Advances in artificial intelligence are rapidly transforming healthcare, offering innovative pathways for non-invasive and affordable diagnostic solutions. In this study, we propose a DL Framework that utilizes facial features for disease prediction, combining theory research with practical system implementation. The approach leverages transfer learning from established facial recognition networks, employing pre-trained convolutional architectures such as ResNet-50, VGG-16, and Inception V3 to process and classify two-dimensional facial images. The pipeline incorporates preprocessing operations—face detection, alignment, normalization, and augmentation—before applying deep transfer learning for classification. Evaluation results indicate that the system achieves accuracy rates above 90% in detecting conditions including beta-thalassemia, hyperthyroidism, Down syndrome, leprosy, and healthy, surpassing traditional machine learning baselines and, in certain cases, clinical assessments.

To validate the framework in practice, we developed a prototype application named EasyFace. This web-based tool allows users to upload facial photographs and receive immediate diagnostic predictions. To enhance transparency, Grad-CAM visualizations are integrated to highlight facial regions most influential in the model’s decision-making process. The system is powered by a lightweight Flask backend, supported by SQLite for persistence, and designed with a user-friendly interface featuring modern animations, error handling, and optional history tracking through JWT authentication. Despite promising results, challenges remain, particularly the limited availability of large, diverse facial diagnosis datasets and the need for continuous refinement of model calibration and interpretability.

This research contributes to the growing field of computer-aided diagnosis by demonstrating how transfer learning can be adapted from facial recognition to medical screening. The outcomes emphasize the potential of facial analysis to support early detection, improve patient care, and expand access to diagnostic services. Future work will focus on extending coverage to additional diseases, incorporating larger datasets, benchmarking against advanced deep learning architectures, and preparing the system for secure, production-level deployment. Collectively, the findings highlight the promise of deep learning–based facial diagnosis as a transformative tool in modern healthcare.