Temporal bone radiology report classification using open source machine learning and natural …

We developed a multi-label classification pipeline, illustrated in Fig. 1, to classify input radiologist reports as normal or abnormal for each of four ear anatomical regions: inner, middle, outer and mastoid. Each report is pre-processed to extract the findings and impression sections and to perform text normalization. The normalized sections are then converted to a discrete numerical feature vector (FV). The FV is input to the four separate machine learning classifiers that label the report normal or abnormal relative to a specific ear region. The models are made accessible to client applications via a web service. Fig. 1Web service and classification pipeline architecture. Client requests include radiology reports that are first normalized and then classified by four region specific models. Label values are returned to client via an HTTP response Our…


Link to Full Article: Temporal bone radiology report classification using open source machine learning and natural …