INTEGRATING the use of artificial intelligence (AI) in the medical field is not easy: the reliability, clinical experience and precision of the human eye are, to date, irreplaceable. There are, however, some very long procedures to be performed manually, and in which help from artificial intelligence would allow to lighten the workload on doctors, on the one hand, and to speed up the start of therapies, on the other. . An example? Evaluation of the tissues to be treated with radiotherapy. To this end, a team of doctors and researchers from the Artificial Intelligence in Medicine Program of Mass General Brigham (UK) worked to create a reliable algorithm. According to what they tell on the pages of Lancet Digital Health, the experts have developed an artificial intelligence capable of identifying and delineating (“segment” in jargon) in a few seconds a non-small cell lung tumor (Nsclc) on images of a CT scan. The performance would be comparable in terms of quality and reliability of the diagnosis to those obtained by doctors, but the algorithm allows to reach the goal in 65% less time.
The importance of speeding up therapy planning
The effectiveness of radiotherapy as the main tool for treating some forms of cancer, including non-small cell lung cancer (NSCLC), depends very much on the timeliness with which treatment is started and on its adequate design. Basically, it depends on how carefully the tissues to be treated are chosen and healthy ones are excluded. A process called segmentation is used to select the tissues invaded by cancer cells and the involved lymph nodes, which requires a thorough analysis of the three-dimensional images collected with the CT scan to decide the target area. The risk, if the segmentation process leaves out already damaged regions, is that the tumor grows in an uncontrolled way; on the contrary, that is, if the selection of the tissues is too generous, the risk is to greatly increase the toxicity of the radiotherapy treatment. Another problem, then, is subjectivity: several studies have shown how the variability of image evaluation by different experts affects the outcome of therapies.
Lung cancer, the Italian project on artificial intelligence and big data is underway
by Barbara Orrico
The help of artificial intelligence
The deep learning algorithm developed in the British study was designed, first of all, to help the clinical evaluation of the images and to speed up the segmentation process of the regions to be treated with therapy. He was then trained to distinguish cancer cells from those of other tissues using computed tomography images of 787 patients. Scientists then tested its performance on scans of 1,421 patients with various tumor types, recalibrating the data set needed to properly train the algorithm based on its performance on various cell types and tissues. Finally, the researchers asked eight radiological oncologists to perform the segmentation and evaluate – without knowing a priori the origin – those produced by another expert doctor or by the algorithm.
Fast and quality support
Analyzing the results of the segmentation operated by the algorithm or manually by the radiologists, the doctors found no difference. Relying on the analysis conducted by artificial intelligence to decide the therapeutic strategy, however, allowed the work to be done 65% faster. Evaluation variability, when checked by multiple expert eyes, also decreased if the starting point was algorithm segmentation: on average, 32% fewer variations were made than manually produced segmentations. The precision in the choice of the tissues to be treated and those to be discarded, and therefore the definition of the outlines of the offended regions, was also higher if drawn with the help of AI.
From prevention to therapy, artificial intelligence revolutionizes the fight against cancer
by Valentina Guglielmo
The results of this study, underline the authors, are important because they demonstrate the possibility and usefulness of a virtuous collaboration between humans and AI in the medical field. The treatment path for lung cancer, for example, involves radiotherapy in almost half of the cases, and currently the identification and planning of the areas to be targeted can take up to several weeks and, considering the increase in the incidence of this and other types of cancer, the times could be lengthened further. Having a faster and more independent method of assessment and diagnosis, therefore, could be valuable in the future and could help even in cases where clinical assessments are uncertain.