About Invention

The present invention relates to a deep learning based method for detecting and segmenting multiple brain anomalies, comprising annotating regions of anomalies on a plurality of original CT scan images; generating a ground truth mask for each of the plurality of annotated images; pre-processing the plurality of the original CT scan images and the plurality of ground truth masks by deducing and further resizing a hounsfield value for each of them; designing a segmentation model based on DeepLabV3/MaNet for detecting and segmenting the multiple brain anomalies, training and validating the segmentation model using a plurality of pre-processed pairs of the original CT scan image and the annotated image thereof; applying the trained segmentation model on one or more user-inputted CT scan images to retrieve a predicted mask; and comparing the predicted mask with the plurality of ground truth masks to detect the brain abnormality.