METHOD FOR DIAGNOSING PROSTATE CANCER USING PROGNOSTIC MODEL USING DEEP LEARNING BASED ON RADIOMIC FEATURES BY INTEGRATED INTERPRETATION OF CLINICAL AND LABORATORY DATA AND bpMRI Russian patent published in 2024 - IPC A61B5/55 G16H30/00 G06T7/00 G06N3/02 

FIELD: medicine.

SUBSTANCE: invention relates to medicine and is intended for diagnosing prostate cancer using a prognostic model using deep learning based on radiomic features by integrated interpretation of clinical and laboratory data and bpMRI. Disclosed is a method for diagnosing prostate cancer using a prognostic model using deep learning based on radiomic features by integrated interpretation of clinical and laboratory data and bpMRI, characterized by the following steps: data collection, data processing, training of the neural network, validation and testing of the neural network, directly the stage of diagnostics (analysis of the obtained images). When collecting data, clinical data of patients with verified pancreatic cancer are obtained, bpMRI images of patients with verified pancreatic cancer are obtained, and total amount of data is distributed into subgroups for training and validation of the neural network. When processing data for subsequent training of a neural network, regions of interest are determined for their subsequent selection based on bpMRI images, region of interest is marked in accordance with the planned functions of the developed neural network, pre-processing of bpMRI images is carried out, after which the previously marked regions of interest are analysed. After marking and preparation of bpMRI images, the neural network is trained, after which the neural network is validated, wherein two groups of cases are formed: training and validation, and each case of both groups includes all selected data, where the validation group includes a set of data necessary to check the operability of the neural network after training using the training dataset. After supplying a series of bpMRI images to the developed system and its segmentation and analysis from a deep artificial neural network, which is used to analyse the image and detect the degree of cancer risk, data are extracted from the output. Array of patient parameters is added to the obtained data, including the PSA density after calculation in the corresponding window and the age taken into account when analysing the images by default. Obtained combined array is transmitted to an additional neural network – a classifier, which, based on the input data, evaluates the risk of the presence of clinically significant forms of cancer in the analysed MRI series.

EFFECT: invention provides higher accuracy and speed of making a diagnosis by a doctor using a prognostic model using deep learning based on radiomic features by integrated interpretation of clinical and laboratory data and bpMRI.

19 cl, 15 dwg