HCC-AI

Detection and delineation of hepatocarcinomas based on AI image analysis.

Hepatocellular carcinoma or hepatocarcinoma (HCC) is a liver cancer that constitutes 80-90% of malignant liver tumours. According to World Health Organisation statistics, it is currently the sixth most prevalent cancer in the world, but is the third leading cause of cancer deaths.

90% of HCCs occur in patients with liver cirrhosis (regardless of the aetiology of the cirrhosis) and curative treatments are available at early stages, ranging from surgical resection to liver transplantation [….]. For this reason, early diagnosis is very important and all scientific societies around the world recommend screening [2][3]. This screening is performed by liver ultrasound at 6-month intervals. If a suspicious lesion is observed, a CT scan or MRI should be performed and if unclear, a biopsy.

Ultrasound is a convenient, inexpensive, minimally invasive method and has a sensitivity of 65-80 % and a specificity of over 90 %, so that sometimes the detection of this tumour is with a large size, or differential diagnosis with other benign tumour lesions can be difficult.

The aim of this project is to develop a model for the detection and delimitation of this type of cancer based on images taken by ultrasound. The model will not only be able to indicate a positive or negative diagnosis, but will also be able to indicate its grade and evolution, as well as to delimit within the images which is the affected area of the liver. If we are able to achieve with this model a high sensitivity and specificity (as well as high positive and negative predictive values) with the ultrasound images, we will be able to have a high diagnostic certainty without resorting to other additional tests or the need for a biopsy of the lesion.

Artificial intelligence techniques will be applied to make this diagnosis, either by developing our own models or by adapting detection, segmentation and classification models developed by third parties whose licence allows them to be used and fine-tuned for a specific case. Once the model has been developed, it will be desirable to be able to evaluate it by developing an application, which could be a web application or a desktop application.

Finally, this project will be carried out in collaboration with professionals from the Hospital Rio Hortega in Valladolid, who will provide labelled images of the different types of hepatocarcinoma and will also provide feedback on the results obtained by the model and the associated application.


TOP