Exposure Fusion Framework in Deep Learning-Based Radiology Report Generator

Writing a radiology report is time-consuming and requires experienced radiologists. Hence technology that could generate an automatic would be beneficial. The key problem in developing automated report-generating system providing coherent predictive text. To accomplish this, it important to ensure the image has good quality so model can learn parts of interpreting, especially medical images tend noise-prone acquisition process. This research uses Exposure Fusion Framework method enhance increase performance producing used encoder-decoder with visual feature extraction using pre- trained ChexNet, Bidirectional Encoder Representation from Transformer (BERT) embedding for text feature, Long-short Term Memory (LSTM) as decoder. model’s EFF enhancement obtained 7% better result than without processing evaluation value Bilingual Evaluation Understudy (BLEU) n-gram 4. It concluded effectively increases performance.