Deep learning is a subfield of artificial intelligence that uses the same network of simple nonlinear functions for both feature extraction and decision-making. Unlike most classical models, the whole network is trained end to end to perform the desired task.
The introduction of deep models has revolutionized the fields of image analysis, speech recognition and language processing and they play a vital role in recent advances towards self-driving cars. The deep model architecture that has recently become the method of choice for image analysis applications is convolutional neural networks (CNN). RECOMIA uses such models for automated analysis of medical images.
The automated organ segmentation is performed using a CNN that has been trained on manually segmented images. Approximately 10,000 manual organ segmentations have been used to train the current versions of our tools. We continue to train new CNNs to be able to continuously improve the performance. RECOMIA’s AI-based tools have showed to provide highly accurate and reproducible results, similar to those obtained manually by radiologists, but much faster.
After automated segmentation of, for example, organs of interest in CT images or tumour tissue in PET images, different measurements can be calculated. Volume of organs, average Hounsfield units inside organs, total lesion uptake (product mean SUV x tumour volume), fraction of organ involved with tumour and total tumour burden are some example of measurements that can be calculated. One or more of these measurements can emerge as a clinically important imaging biomarker.
RECOMIA´s AI-based tools can segment 100 organs and volumes:
CT - Bones
Cervical vertebra all
Thoracic vertebra (Th1, Th2, …, Th12)
Ribs (Left1, right1, left2, …. , right 12)
Sternum (body and manubrium)
Lumbar vertebra (L1, L2, …, L5)
Hip bone (left/right)
Sacrum and coccyx
CT - Soft tissue organs
Aorta thoracic part
Aorta abdominal part
Iliaca communis (left/right)
Adrenal glands (left/right)
Gluteus maximus muscle (left/right)
High uptake in Lymph nodes
High uptake in Urinary bladder
High uptake in Bones
High uptake in Prostate
Prostate peripheral zone
Prostate central/trans zones