Artificial intelligence has the potential to improve medical imaging data analysis. For example, deep learning-based algorithms can determine the location and size of a tumor. This is the result of Karlsruhe Institute of Technology (KIT) researchers taking 5th place in AutoPET, an international medical image analysis competition. Seven of the best autoPET teams report in the journal Nature Machine Intelligence on how their algorithms can detect tumor lesions in positron emission tomography (PET) and computed tomography (CT).
Imaging technology plays an important role in cancer diagnosis. To choose the right treatment, it is essential to accurately determine the location, size, and type of the tumor. The most important imaging techniques include positron emission tomography (PET) and computed tomography (CT). PET uses radionuclides to visualize metabolic processes in the body. The metabolic rate of malignant tumors is significantly higher than that of benign tissue. Radiolabeled glucose, typically fluorine-18-deoxyglucose (FDG), is used for this purpose. In CT, the body is scanned layer by layer through an X-ray tube to visualize anatomy and locate tumors.
Automation saves time and improves assessments.
Cancer patients sometimes experience hundreds of lesions, or pathological changes resulting from tumor growth. To obtain a uniform picture, all lesions must be captured. The physician determines the size of the tumor lesion by manually marking the 2D slice images. This is a very time-consuming task. “Automatic evaluation using algorithms will save enormous time and improve results,” explains Professor Rainer Stiefelhagen, Head of Computer Vision for KIT’s Human-Computer Interaction Laboratory (cv:hci).
Rainer Stiefelhagen and Zdravko Marinov, PhD students at cv:hci, participated in the 2022 International autoPET Competition and placed 5th out of 27 teams with 359 participants from around the world. The Karlsruhe researchers teamed up with professors Jens Kleesiek and Lars Heiliger from the Essen-based Institute for Artificial Intelligence in Medicine (IKIM). Organized by the University Hospital Tübingen and the LMU Hospital Munich, autoPET combines imaging and machine learning. The task was to automatically segment metabolically active tumor lesions visualized on whole-body PET/CT. For algorithm training, participating teams had access to large annotated PET/CT datasets. All algorithms submitted to the final stage of the competition are based on deep learning methods. This is a variation of machine learning that uses multilayer artificial neural networks to recognize complex patterns and correlations in large amounts of data. The top seven teams in the autoPET competition have now reported on the potential for automated analysis of medical imaging data. natural machine intelligence newspaper.
Algorithm ensemble superior for tumor lesion detection
As the researchers explain in their publication, ensembles of top-rated algorithms have proven to be superior to individual algorithms. The ensemble of algorithms can efficiently and accurately detect tumor lesions. “The performance of an algorithm in the evaluation of image data depends in part on the quantity and quality of the data, but algorithm design is another important factor, for example in relation to decisions made in post-processing of predicted segmentation.” Stiefelhagen explains. Additional research is needed to improve the algorithm and increase its resistance to external influences so that it can be used in routine clinical practice. The goal is to fully automate medical PET and CT image data analysis in the near future.
