Cancer is a leading cause of death, with 9.7 million cancer-related deaths in 2022.
New cases are expected to climb from 20 million in 2022 to 29.9 million in 2040 [1]. With such grim statistics, there are concerted efforts to tackle the disease, ranging from precision medicine to drug development. Multiple innovations and technologies are helping in this fight, with artificial intelligence(AI) playing a significant role.
AI holds immense potential to revolutionize cancer treatment on a global scale. Beyond improving therapies, AI-powered systems promise to reduce cancer disparities by enhancing access to optimal treatment and bridging the gap in cancer care and research.
In a recent podcast called “Let’s Talk Cancer”, Dr. Laszlo Radvanyi, President and scientific director of the Ontario Institute for Cancer Research, admits that AI has revolutionized various aspects of cancer treatment, and much more is expected in the future [2].
These sentiments show AI's potential in the fight against cancer. This article explores how AI impacts cancer treatment in areas like early detection, personalized patient care, and research.
"Diagnosing cancerous lesions is a laborious and challenging task. We want every lab in the world to have access to our AI application. It can triage samples, provide a second opinion, or even provide a preliminary diagnosis in underserved regions," says Dr. Chao, a lead researcher developing AI-based programs for oral cancer. This sentiment echoes the expectations of many researchers and medical professionals regarding the potential of artificial intelligence (AI) in healthcare.
The explosion of electronic healthcare data has created vast datasets that can be analyzed using AI techniques, such as machine learning (ML) and deep learning (DL). These powerful algorithms can uncover patterns and correlations within the data that may not be apparent through traditional analysis methods. ML and DL techniques are particularly effective in cancer detection and diagnosis, as they can quickly process large amounts of data and identify subtle indicators of the disease.
Identifying people with a high risk of suffering from cancer is advantageous since it allows doctors to monitor them and deploy early interventions. Medical imaging techniques like PET scans, MRI, and CT scans provide the basis of image evaluation. However, interpreting some of these images is complex due to different tumor presentations and other overlapping features between malignant and benign lesions.
DL algorithms like generative adversarial networks(GANs) and convolution neural networks (CNNs) can extract relevant information from scans [3]. AI-powered risk prediction models can use multi-modal data to determine the likelihood of a patient developing cancer based on risk scores.
Research from the Radiological Society of North America (RSNA) determined that AI models perform better than standard risk models like BCSC at predicting the five-year risk for breast cancer. The AI-based tools use mammograms and can identify missed cancers and breast cancer tissue [4].
AI-based image analysis is also used to develop tools to screen and predict people likely to have lung cancer. For instance, researchers have developed a tool known as Sybil, which can accurately predict lung cancer using a single low-dose chest CT scan. The tool can do this for short and long term, simplifying risk assessment processes and helping to improve screening [5].
There is also AI-based softwares designed to help pathologists to screen and diagnose prostate cancer. The system analyzes prostate biopsy images to identify whether the image shows areas with cancerous tissues[6].
Additionally, AI-driven digital pathology enhances cancer diagnostics using algorithms like artificial neural networks(ANN). The system studies and identifies the development of novel biomarkers from high-resolution tumor images [7].
Another tool called iStar interprets medical images in detail, providing clinicians with clarity on cancers that might have been missed. Additionally, it can determine the safe margins during cancer surgeries and label microscopic images automatically, which helps in molecular disease diagnosis [8].
These simplified, noninvasive screening processes show promise in promoting the fight against cancer across different geographic and demographic contexts. The use of medical data and images to anticipate and diagnose disease will continue to develop as more AI-based tools become available.
Some advanced AI tools are in development to analyze blood profiles, specifically examining circular tumor DNA(ctDNA) and microRNA(miRNA) in plasma [9]. These can be used in detecting pancreatic and gastric cancers, which are very challenging to detect.
The shift towards AI-driven blood profiling promises earlier and more accurate cancer detection. For instance, a novel technique by Johns Hopkins Kimmel Cancer Center showed 91% to 96% accuracy in detecting lung cancer through blood samples [10].
Numerous AI-powered tools, apps, and devices promote preliminary screening for early detection. However, they do not replace medical evaluation and expertise.
A good example is an app called SkinVision, a regulated medical device that screens skin for cancer. The app allows users to take photos of suspicious skin spots. The AI analyzes the image to identify the skin type and risk profile. The tool has an accuracy rate of 95% for common skin cancers [11].
One of the most significant challenges with conventional methods is that they focus on killing as many cells as possible. These methods subject patients to high doses, which may not always have the desired curative effect. Due to cancer’s high heterogeneity, similar medications may not work the same in different patients.
ML and DL models can analyze a patient's deep-level genomics, proteomics, pathological images, and other data to create a comprehensive patient profile [12].
In a study by the University of Oxford and Moffitt Cancer Center, researchers successfully used a deep-reinforcement learning(DRL) framework to develop a personalized cancer care approach. This method allowed medics to formulate individualized treatment plans for prostate cancer patients, helping to double the time to relapse for some patients [13].
Using this framework, the researchers proposed a new method to create a “virtual twin” of the patients. The DRL framework could also be finetuned to produce a personalized treatment schedule for new cancer patients based on their initial treatment data [14].
Treatment of patients with cancers of unknown primary(CUP) is challenging because their origin is unknown. AI is also helping researchers make significant strides in this area, assisting doctors in personalizing treatment plans based on their outcomes while avoiding broad-based drugs. The tool analyzes the genetic sequences of 400 cancer-related genes to predict a tumor's origin, which is critical to determining an effective treatment strategy [15].
The oncoNPC model used in this study helped double the number of patients eligible for genomically guided treatments. Oncologists can prescribe precision drugs according to the predicted cancer type, improving patient outcomes and reducing side effects from broad-spectrum drugs [16].
There’s normally reduced cancer drug efficacy after the initial period of treatment due to drug resistance. However, research has shown that combination therapy has superior efficacy, reducing drug resistance and dosage. Studies have also indicated higher survival rates for patients receiving combination therapies than those undergoing single-agent treatment.
ML and DL models using algorithms like XGBoost, feed-forward neural networks, and random forest(RF) can be used for high-throughput screening to accelerate the classification and identification of anti-cancer drug synergy [17]. Researchers can develop rational drug combinations for effective therapies by predicting novel synergistic combinations.
AI-powered research and development in cancer care is remarkable, underscoring extensive studies and groundbreaking discoveries. AI has continued to present promising methods to advance cancer research, especially when integrated with other cutting-edge technologies such as radionics and genomics [18].
Developing anti-cancer medicine is expensive, complex, and time-consuming. With the increase in cancer cases globally, it has become urgent to speed up the development of effective drugs to combat the disease.
Here is how AI is helping in cancer drug development.
Traditionally, tumor and organ segmentation requires considerable time from radiation oncologists. It's a complex workflow that involves modeling the patient, simulations, planning, and delivering treatment. Radiologists contour the target of the organs and tumors at risk, which then are dosed with radiation.
The use of AI has enhanced different aspects of cancer radiation therapy, including:
Reducing the time needed for treatment planning and optimization from days to minutes. For instance, Siemens Healthineers is developing an AI-powered system for oncologists to develop treatment plans faster. The AI-assisted AutoContouring was trained with over 4.5 million images and can automatically outline 47 organs [21].
AI also promises to boost quality control, optimize image-guided radiation therapy, and monitor mobile tumors during treatment [22].
AI is expected to significantly impact cancer treatment, improving numerous applications. However, the use of AI in cancer detection and treatment still faces several challenges.
Data quality: The data used to train the model lacks standards, is imbalanced, and is disordered because it originates from different systems. This makes it difficult to build models with high precision. For instance, devices that generate pathological images accept different parameters, so the image data differs. Other challenges include insufficient training data and data management.
Ethical concerns around privacy: Patient data is governed by privacy laws. Lack of patient consent and data mishandling may lead to patient privacy violations. For instance, Google and the University of Chicago were involved in a lawsuit because the University shared identifiable medical data with the tech firm [23].
Lack of explainability: It becomes difficult for researchers and medics to explain the outcome of AI systems, especially for complex models. If the system does not reveal its reasoning, it becomes difficult to precisely determine how the model arrived at its result, leading to mistrust. This limitation can be tackled by introducing a layer of explainable AI, boosting transparency and better interpretability of outcomes.
Regulatory framework: Questions abound about whether AI in health fits into the current legal frameworks or whether specific legislation is needed. Although the USA and Europe have instituted various laws and policies to regulate the use of AI, there is still a need to consolidate this and clarify various aspects. For instance, who is the data owner if an AI model generates new data from existing medical records? Is there a need to seek additional consent in this case?
The global healthcare community recognizes the immense challenges posed by cancer, as the number of cases and deaths continues to rise. Traditional workflows often struggle to address the complex aspects of this disease effectively. However, the emergence and growth of technologies like artificial intelligence (AI) provide new avenues to tackle cancer.
AI has shown significant potential in cancer detection and treatment, with extensive research and development efforts underway. As the availability of medical data increases, numerous studies are being conducted, leading to breakthroughs in AI-assisted cancer detection, drug development, and research.
Despite the progress, challenges such as data bias and regulatory hurdles hinder the widespread adoption of AI in cancer treatment. As we move forward, we will likely witness further advancements in AI-based healthcare applications and a streamlining of regulations to overcome initial obstacles. The future holds promise for AI to revolutionize cancer care, ultimately improving patient outcomes and reducing the global burden of this devastating disease.
