Automatic assignment of diagnosis codes to free-form text medical note

International Classification of Disease (ICD) coding plays a significant role in classify-ing morbidity and mortality rates. Currently, ICD codes are assigned to patient’s medical record by hand practitioners or specialist clinical coders. This practice is prone errors, training skilled coders requires time human resources. Automatic prediction can help alleviate this burden. In paper, we propose transformer-based architecture with label-wise attention for predicting on dataset. The transformer model first pre-trained from scratch Once done, the used generate representations tokens documents, which fed into layer. Finally, outputs layer feed-forward neural network predict appropriate input document. We evaluate our using hospital discharge summaries their corresponding ICD-9 MIMIC-III Our experimental results show that outperforms all previous models terms micro-F1 full label set also successful application auto-coding problem