A Hybrid Approach Combining YOLO and Deep Neural Networks for Efficient Lung Disease Detection

Lung diseases are a main source of concern for morbidity and mortality globally; thus, early diagnosis plays a great role in improving the prognosis for patients. Chest X-ray imaging which is a common imaging modality for diagnosing lung diseases usually requires image interpretation an aspect that can be hard in areas of limited resources. This research outlines a fusion approach of YOLO for real time object identification and Deep Neural Network classifier for accurate classification of the diseases. By using the NIH Chest X-ray Images Dataset of 112120 labelled images of 14 thoracic conditions the model proposes improving both the localisation and classification results. YOLO identifies pathologies from chest X-rays and visualizes bounding boxes with confidence score and the DNN differentiates these areas int]] specific diseases. Having tested the system, it is shown that the proposed hybrid model of YOLO and the DNN is more accurate, with a record accuracy of 96.5% compared to the standalone accuracies of YOLO and DNN which were 92.1% and 93.8% respectively. Reinforcement of the observation is the fact that the capacity of the model to perform an inference takes 25 ms per image, thus it is real-time applicable. The representation of qualitative visual analysis also induces a sense of confidence for the correct identification and categorization with complicated instances. Some uses involve using them in the patient care environments to support radiologists and in telemedicine environments to perform diagnoses. The future work includes the research of integration of multi-modal data, enhancing the model for the edge devises and its generalization to the different people. In the proposed approach, it is possible to maintain a high level of reliability and scalability of the further development of automated diagnostic tools in the field of healthcare.