Cancer’s Greatest Weekness – Early Detection and Treatment
Cancer is a leading cause of death globally. The physical, emotional, and financial burden of cancer, exerts tremendous strain on individuals, families, and healthcare systems. Many types of cancers can be treated and healed early in their growth only if they are detected early.
Accurate early diagnosis and predicting the ideal course of medication is critical to improving the patient’s survival rate. The traditional method of a pathologist manually looking at a histology slide to analyse over half a million cells to share an accurate diagnosis is extremely challenging and more importantly, the chances of misdiagnosis or late diagnosis is higher. And no one would want a high false-positive or false-negative result while diagnosing any disease especially, cancer!
So, how can AI help in diagnosing cancer?
AI finds what eludes the human eye. It increases the precision in diagnosing cancer by recognizing even the most subtle disease patterns within the patient’s cells, which is difficult for the human eye to detect. This can guide physicians and pathologists in diagnosing cancer more accurately and provide better-targeted therapies and improved patient outcomes.
In the diagnosis of many cancers, including those that occur in the brain, such as medulloblastoma and glioblastoma, a full-genome methylation analysis is often needed. This is used to check for small hydrocarbon molecules attached to DNA that can alter chromosomes’ activity and cause different types of cancers. Interestingly, when these methylation analysis results are fed to an artificial intelligence (AI) system, the machine efficiently classifies the tumors based on the different patterns of methylation. The computer’s ability to spot these cancer types can also cut hospitals’ error rates.
The German Cancer Research Center in Heidelberg and NYU Langone’s Perlmutter Cancer Center are among the pioneers who have made AI-based “classifiers” available online so that researchers can upload the methylation profiles to correctly identify the subtype of cancer.
“Although the classifier is still a research tool that has not been clinically validated, it is trained using around 60,000 tumour samples which is much more than what a single pathologist sees in an entire lifetime,” says Andreas von Deimling, a neuropathologist at the German Cancer Research Centre.
FocalNet is such an AI-based tool developed by UCLA that helps physicians classify prostate cancer better. Physicians usually use MRI to diagnose prostate cancer, which requires a lot of training and several years of experience. So, when a subset of the scans is fed on FocalNet, along with the tumor’s rating, the system looks for patterns in the MRI scans that match the pathology-based score.
Similarly, AI-based automated pathology labs to detect many types of cancer are becoming a reality in today’s world. US-based DeepLens is one of the world’s first digital pathology that provides AI-powered image detection and workflow support, telepathology, collaboration, and cloud storage. Israel-based Ibex also uses AI-based algorithms to analyze images, detect and grade cancer in biopsies, helping pathologists reduce diagnostic error rates and enable a more efficient workflow.
Getting it right – False Positives and Negatives.
False positives and false negatives in diagnosing any kind of cancer can affect the survival rate and prognosis. This can either lead to treatment that is delayed or unwanted. For example, a patient with breast cancer who has to pursue significant surgeries, like double mastectomies, when they may not require or failure in cancer detection, could cause cancer to spread onto other parts.
When the machine-learning algorithm was used by scientists at MIT for a subset of mammogram images, 31% of the women were put into the highest risk group, who could eventually develop breast cancer. In contrast, the physicians’ traditional model identified only 18% in the highest risk group. Such innovative findings can lead to more personalized breast-cancer screening and offer mammograms only to those whose early scans show they are at high risk.
France based start-up Primaa, uses a deep-learning algorithm to assist anatomical pathologists, particularly in breast cancer detection, while Paige.AI is another new start-up that focuses on breast, prostate, and other primary cancers. Paige is being built by a group of experts in pathology, machine learning, healthcare, and institutions like Arterys, Memorial Sloan Kettering Cancer Center, Merck, NASA, and NVIDIA.
Researchers at Google, using AI to analyze low-dose CT scans of the lungs, found that the machine learning algorithm was more accurate than radiologists at correctly identifying tumors and malignant growths in scans, even for people that had only one scan. AI boosted diagnosis is being extended in other types of carcinomas too.
UK-based Skin Analytics uses artificial intelligence to screen for melanoma skin cancer, while Japan-based AI Medical Service developed an AI software to help detect gastric cancer. Endofotonics that is based in Singapore focuses on SPECTRA IMDx system that enables AI-enabled real-time detection of early gastric cancer during endoscopy.