AI systems vs. cancer: diagnosis, treatment, follow-up

AI systems vs. cancer: diagnosis, treatment, follow-up
© iStock/4X-image

From surgery assistance to virtual nursing, next-level AI systems help healthcare providers to facilitate and accelerate cancer diagnostics and treatment response.

As global computational powers keep growing, artificial intelligence (AI) is getting embraced in healthcare more and more. From surgery assistance to virtual nursing, AI systems gain higher trust among providers and unlock new opportunities for supporting diagnosis and clinical decision-making. For example, Signify Research forecasts a wave of AI investments in medical image analysis with about $2 billion (~€1.8 billion) in annual spending by 2023.

The R & D department of custom healthcare software company Itransition comments on the prospects of AI evolution: “As artificial intelligence technology wins credibility and matures further, we expect its wider adoption in research and clinical activities. In a decade, AI will be advanced enough to stand at the forefront of healthcare, tackling the most pressing challenges we face now. Think genome sequencing, outbreak prevention and personalised care delivery for patients with complex or rare conditions.”

Certainly, cancer is one of such conditions, being incredibly varied with over 200 types and different timelines for development, growth, and spread. According to the WHO, cancer remained one of the leading causes of death globally in 2018, and was responsible for about 9.6 million deaths.

Facilitating and accelerating cancer diagnostics are among the primary goals to achieve in oncology. This will allow for timely intervention and more gentle therapy choices for all types and forms of cancer. AI can be just the right approach to meet this goal, also contributing to the cancer care cycle by helping providers to balance the therapy and track the treatment response.

Finding the big C

Depending on the AI system configuration and maturity level available for a particular provider, the algorithms can bring benefits to cancer diagnostics in several ways. As it processes massive amounts of data in a split second, AI can cut time to diagnosis by analysing medical images, finding the lesions with high accuracy and presenting the findings to the pathologist.

Multiple studies prove that the current AI is able to match and surpass the human eye in detecting abnormalities. For example, a recent competition in localising brain tumours and predicting hematoma expansion in Beijing wrapped up with the BioMind AI System scoring a 2:0 victory against human pathologists in both rounds. The system achieved 87% accuracy in diagnosing brain lesions across 225 cases in 15 minutes. It took the team of 15 top-level health specialists 30 minutes to achieve 66% accuracy.

BioMind also hit 83% accuracy in predicting brain hematoma expansion compared to 63% from the physicians’ team. It’s especially important to note that the system’s win doesn’t mean we won’t need human health specialists soon enough. It just shows that medical specialists can entrust the algorithms with doing routine and preliminary job. This would free specialists’ time to interact with their patients face to face and tailor the treatment plan.

Other scientists targeted early detection of colorectal cancer. This cancer starts from benign polyps forming in the colon, and it may take years for them to eventually become malignant. Even then, the symptoms can appear too subtle or vague to recognise cancer early, such as cramps or change in the bowel movement (e.g. diarrhoea or constipation).

The researchers addressed the problem and created an algorithm to detect colorectal cancer early, achieving 86% accuracy.

Google’s researchers headed in a slightly different direction, employing AI for detecting metastases in breast cancer cases. According to the study, current metastasis evaluation techniques lack accuracy, which can lead to changes in staging upon the second pathology review. In turn, it can significantly change treatment decisions and hinder patient care, especially if smaller metastases weren’t noticed timely.

Their LYmph Node Assistant, or LYNA, achieved 99% accuracy in classifying metastatic breast cancer. This showed the algorithm’s potential to help pathologists avoid double-guessing in detecting even smaller cancer spread areas and as well as help them make the most patient-beneficial therapy choices.

As powerful as it is, AI can bring even more value as it matures. If used for routine cancer screenings, AI systems can accumulate scanning results for particular patients over time instead of giving out an instant suggestion for a single session. Accessing the imaging history for individual patients, AI can analyse oncologic biomarkers and put them into the perspective, alerting the specialists about suspicious patterns long before they become noticeable by the human eye.

Taking it up a notch, mature AI can process a patient’s lab results and other medical data from their history to correlate their life habits and health risks with the screening results. This would help to initiate an array of preventive measures or offer additional tests for cancer detection as early as on a cellular level.

Tuning cancer therapy

Proceeding with therapy, the role of AI role is no smaller. When it incorporates novel clinical trials and recent breakthroughs, it enriches human pathologists’ expertise and helps them generate incredibly personalised patient treatment options with prognosed symptoms and health outcomes.

For example, a recent UK-US research has shown how AI can use patient data to simulate different treatment responses and predict therapy resistance development with further cancer evolution. Using this information, health specialists can discard obviously ineffective medications and create highly targeted treatment plans.

According to the MIT scientists, next levels of AI maturity can make the algorithms fine- tune the dosing to help patients achieve the best outcomes with minimised discomfort and adverse effects. They have tested a reinforced learning (RL) model to reduce chemotherapy and radiotherapy dosages for 50 simulated patients with glioblastoma, helping to shrink the tumour as fast as possible but in a most gentle way. They achieved significant cuts on treatment dosages and frequency of their administration, thus reducing the overall therapy toxicity while preserving its efficiency.

Such adverse effects as nausea, lack of appetite, fatigue, numbness, and hot flashes can be common in patients under chemotherapy. The researchers from the University of Surrey and the University of California employed AI to find the correlations between these and other 33 symptoms in 1,300 patients. They hope that the study findings will help health specialists navigate the therapy course in a more controlled way, fine-tuning symptom management both for curative and palliative treatment cases. Additionally, such studies can serve as a basis for thorough patient education, making the therapy more transparent, predictable and tolerable for patients, and helping them cope with their conditions better.

Evaluating treatment results

When all the shots die down and the powder haze dissipates on the big С battlefield, a mature enough AI system can measure the treatment success. Applied in a similar way to the cancer diagnosis logic, the system can process a set of patient scans across different stages of the treatment. While some cases can be quite easy to handle by humans, AI can do a comparative analysis to assess postoperative and post-therapy results at a micro scale.

The system’s sensitivity can allow tracking a patient’s progress at early cancer stages, grasping both negative and positive patterns faster and with higher accuracy. This can help pathologists to make informed decisions about further actions, if needed.

Sometimes the treatment is administered to shrink the tumour and make it operable, other cases require to evaluate an inflammation change around the lesion to choose the next care steps. For follow-ups, the significance of AI can fluctuate from “just wanted to make sure” to “we need to make the best possible decision.” It’s good to be able to use it anyways.

In AI we trust

The main reason for using AI in cancer-related efforts is that it greatly speeds up time. AI algorithms can cut the time needed to diagnose a patient or plan a balanced therapy and add it where needed most. With its help, health specialists can focus on their patients and potentially improve the efficiency of treatment.

Next-level AI systems can do even better by translating multiple medical images into insights about patient health patterns and revealing the risks while they are still reversible or can be mitigated without making major concessions. Regardless of the cancer type or stage, artificial intelligence offers cancer patients more time – to enter remission, to remain strong and capable, and to live at the fullest.

Inga Shugalo
Healthcare Industry Analyst at Itransition

Subscribe to our newsletter


Please enter your comment!
Please enter your name here