Use Deep Learning for Processing Automation Image DR in Detecting Pneumothorax

Abstract

Pneumothorax is condition medical serious thing that happened consequence accumulation air in the pleural cavity, which can cause the collapse lungs and potentially threaten soul If No quick handled. A quick and accurate diagnosis is essential. For determine action proper medical. Digital radiography (DR) is one of the method the most common imaging used in detect pneumothorax. However, the limitations in manual interpretation by manpower medical can cause misdiagnosis or​ delay in handling. Study This propose approach based on Deep Learning, especially Convolutional Neural Networks (CNN), for automation processing DR image in detect pneumothorax. The model used utilise ResNet-50 and DenseNet-121 architectures with transfer learning techniques for increase accuracy classification. The data used originate from the ChestX-ray14 and SIIM-ACR Pneumothorax Challenge datasets that have been annotated by experts radiology. Research result show that the CNN model was developed reach level accuracy of 92%, with a precision of 90%, a recall of 93%, and an F1-score of 91%. In addition, the technique Grad-CAM visualization is used For increase interpretability of the model with highlight important areas in the image that becomes base decision classification. Implementation of this model No only increase efficiency of pneumothorax diagnosis but can also reduce burden Work power medical as well as increase quality service health . With promising results , research​ This open opportunity For development more carry on in application of AI in the field radiology.