PING AN TECHNOLOGY (SHENZHEN) CO., LTD.

A method for detecting and locating a lesion in a medical image is provided. A target medical image of a lesion is obtained and input into a deep learning model to obtain a target sequence. A first feature map output from the last convolution layer in the deep learning model is extracted. A weight value of each network unit corresponding to each preset lesion type in a fully connected layer is extracted. For each preset lesion type, a fusion feature map is calculated according to the first feature map and the corresponding weight value and resampled to the size of the target medical image to generate a generic activation map. The maximum connected area in each generic activation map is determined, and a mark border surrounding the maximum connected area is created. A mark border corresponding to each preset lesion type is added to the target medical image.

The application discloses a main-machine interaction method and system, a computer device, and a storage medium. The method includes: an interaction end collects user information including biological sample information and preprocesses and filters the user information, and sends the preprocessed information to a central server; the central server assigns a biological recognition task and distributes the biological recognition task to a computing server; the computing server extracts biological characteristics from the biological sample information according to the biological recognition task, and returns obtained extraction results to the central server; the central server combines the extraction results to get a processing result; the processing result is returned to the interaction end; the interaction end responds to the processing result to complete a man-machine interaction with the user.

A long short-term memory (LSTM) model-based disease prediction method and apparatus, a computer device, and a storage medium are provided. The method includes: obtaining first medical data of a target object and second medical data of an associated object; inputting the first medical data and the second medical data into a first LSTM network in the LSTM model, to obtain a hidden state vector sequence in the first LSTM network; inputting the hidden state vector sequence into a second LSTM network for operation, to obtain a disease prediction result; selecting a predicted disease with an incidence rate higher than a preset threshold, and recording the predicted disease as a designated disease, and obtaining, based on a preset disease association network, an associated disease directly connected to the designated disease; and outputting the disease prediction result and the associated disease, thereby improving the prediction accuracy.

A long short-term memory (LSTM) model-based disease prediction method and apparatus, a computer device, and a storage medium are provided. The method includes: obtaining first medical data of a target object and second medical data of an associated object; inputting the first medical data and the second medical data into a first LSTM network in the LSTM model, to obtain a hidden state vector sequence in the first LSTM network; inputting the hidden state vector sequence into a second LSTM network for operation, to obtain a disease prediction result; selecting a predicted disease with an incidence rate higher than a preset threshold, and recording the predicted disease as a designated disease, and obtaining, based on a preset disease association network, an associated disease directly connected to the designated disease; and outputting the disease prediction result and the associated disease, thereby improving the prediction accuracy.

A machine learning based medical data classification method is provided. The method includes: a medical data classification request including medical record information is received; a preset medical term base is obtained, and word segmentation is performed on the medical record information according to medical terms in the medical term base to obtain multiple text vectors; features of the multiple text vectors are extracted to obtain multiple text vectors and corresponding feature dimension values; a target classifier is trained with multiple pieces of medical data, and the multiple text vectors and the corresponding feature dimension values are traversed and calculated; until a target node corresponding to the multiple text vectors is traversed, class probabilities corresponding to the multiple text vectors are calculated according to the target node, and a class result corresponding to the medical record information is obtained according to the class probabilities and is pushed to a terminal.

A method for detecting and locating a lesion in a medical image is provided. A target medical image of a lesion is obtained and input into a deep learning model to obtain a target sequence. A first feature map output from the last convolution layer in the deep learning model is extracted. A weight value of each network unit corresponding to each preset lesion type in a fully connected layer is extracted. For each preset lesion type, a fusion feature map is calculated according to the first feature map and the corresponding weight value and resampled to the size of the target medical image to generate a generic activation map. The maximum connected area in each generic activation map is determined, and a mark border surrounding the maximum connected area is created. A mark border corresponding to each preset lesion type is added to the target medical image.

A medical image classification method such as CT (or CAT) scans includes receiving the CT scan or medical image, inputting the medical image into an image classification model, which provides a cross entropy (CE) loss function and an aggregated cross entropy (ACE) loss function. According to the ACE loss function, image samples with generic label are used as input data during model training. The medical image can be classified by using the image classification model, and a classification of the medical image is thereby obtained. The present disclosure can classify indeterminate or general medical images and even unlabeled images and thus realize supervision of medical data. A device for applying the method is also provided.