We present a method for detecting the presence of prostate cancer (PC) in suspicious patients with Clinical Decision Support Systems (CDSS), including training neural networks based on features characteristic of malignant neoplasms, through a comprehensive interpretation of magnetic resonance imaging (MRI) data of the prostate. This study has features that we additionally carry out a comprehensive interpretation of transrectal ultrasound and histoscanning (HS) data.

Between May 2024 and December 2024, fifty patients with suspected PC underwent transrecatal ultrasound, HS, and MRI of prostate and fusion transrectal prostate biopsy. HS and MRI, a layer-by-layer scan of the prostate with verified PC is carried out with the storage of images on a computer in DICOM format, and for the analysis of the obtained data, a deep convolutional neural network with U-Net architecture, an ensemble of neural networks configured to search for large tumors, small anomalies and inflammation, a transfer learning method, a regularization method (Dropout), which randomly turns off neurons during training and an optimization method using the Adam algorithm to reduce overt-training of the neural network are used. Histological results of biopsy obtained from the selected pathological zones in ultrasound examination, HS, and MRI were analyzed. The data from transrectal ultrasound sources, HS, MRI and histological results data are integrated using an ensemble of neural networks, each of which analyzes data from a specific type of scanning while using multitask learning, where one neural network analyzes all three types of images simultaneously. Different layers were used to highlight features characteristic of malignant neoplasms, and the presence of an oncological process in the prostate gland is detected based on the input data of the patient under study.

After the biopsy, the histological analysis results are loaded into the system. Patient M. has verified adenocarcinoma with 7 points on the Gleason scale (3+4).The doctor manually confirms that the biopsy area matches the area allocated by the system. The system updates its algorithms based on the received data for further training, which increases the accuracy of the analysis for future patients. The system for outputting the final report with key findings: • Patient: M., 62 years old. • Suspicious area: right lobe of the prostate, 12×9×8 mm. • Probability of malignancy (according to the SPPVR data): 92%. • Histological conclusion: adenocarcinoma, 7 points on the Gleason scale. • Recommendations: discuss treatment tactics taking into account the stage of the disease (surgery or radiation therapy). The export report is maintained within the information system for discussion at the consultation and further treatment planning.

This method provides an increase in the PC detection under the training of our neural networks including preliminary transrectal ultrasound algorithm, HS, MRI, and histological examination data.