A Data-Efficient Deep Learning Based Smartphone Application For Detection Of Pulmonary Diseases Using Chest X-rays

TL;DR

This paper presents a smartphone-based AI application for pulmonary disease detection from chest X-rays, achieving high accuracy with minimal data, and enabling accessible, real-time healthcare diagnostics.

Contribution

The study introduces a novel mobile app leveraging a robust AI model with data augmentation and semi-live training, improving pulmonary disease diagnosis in low-data scenarios.

Findings

Achieved 99.30% accuracy in binary classification

Achieved 98.40% accuracy in multi-class classification

Demonstrated robustness with minimal and noisy data

Abstract

This paper introduces a paradigm of smartphone application based disease diagnostics that may completely revolutionise the way healthcare services are being provided. Although primarily aimed to assist the problems in rendering the healthcare services during the coronavirus pandemic, the model can also be extended to identify the exact disease that the patient is caught with from a broad spectrum of pulmonary diseases. The app inputs Chest X-Ray images captured from the mobile camera which is then relayed to the AI architecture in a cloud platform, and diagnoses the disease with state of the art accuracy. Doctors with a smartphone can leverage the application to save the considerable time that standard COVID-19 tests take for preliminary diagnosis. The scarcity of training data and class imbalance issues were effectively tackled in our approach by the use of Data Augmentation Generative Adversarial Network (DAGAN) and model architecture based as a Convolutional Siamese Network with attention mechanism. The backend model was tested for robustness us-ing publicly available datasets under two different classification scenarios(Binary/Multiclass) with minimal and noisy data. The model achieved pinnacle testing accuracy of 99.30% and 98.40% on the two respective scenarios, making it completely reliable for its users. On top of that a semi-live training scenario was introduced, which helps improve the app performance over time as data accumulates. Overall, the problems of generalisability of complex models and data inefficiency is tackled through the model architecture. The app based setting with semi live training helps in ease of access to reliable healthcare in the society, as well as help ineffective research of rare diseases in a minimal data setting.