AI can mistakenly see cancer in medical scans after tiny image tweaks

New Scientist

Medical artificial intelligence breaks a little too easily. Although AI promises to improve healthcare by quickly analysing medical scans, there is increasing evidence that it trips up on seemingly innocuous changes. Sam Finlayson at Harvard Medical School and his colleagues fooled three AIs designed for scanning medical images into misclassifying them by simply altering a few pixels. In one example, the team ever so slightly altered a picture of a mole that was first classified as benign with 99 per cent confidence. The AI then classified the altered image as malignant with 100 per cent confidence, despite the two images being indistinguishable to the human eye.

Artificial Intelligence: The New Life Jacket of the Healthcare Space


India has two contrast images when it comes to the healthcare sector. While one faction is able to avail the advancement of technologies in the healthcare space, another faction can at the most only avail the basic healthcare treatments. But this spectrum is altering now. Starting from making the basic healthcare services available to the needy, several innovative technologies are disrupting the space and taking them to an altogether different standard. The private healthcare sector in close association with startups has been seeing the innovation that has amazed everyone and spend the industry up.

Using machine learning for medical solutions


Pharmaceutical companies spend a lot of time testing potential drugs, and they end up wasting much of that effort on candidates that don't pan out. Kyle Swanson wants to change that. A master's student in computer science and engineering, Swanson is working on a project that involves feeding a computer information about chemical compounds that have or have not worked as drugs in the past. From this input, the machine "learns" to predict which kinds of new compounds have the most promise as drug candidates, potentially saving money and time otherwise spent on testing. Several prominent companies have already adopted the software as their new model.

25 DeepTech News Briefs


The Stanford Institute for Human-Centered AI officially launched today. Stanford HAI seeks to become an interdisciplinary global AI hub and to fundamentally change the field of AI by integrating a wide range of disciplines and prioritizing true diversity of thought. Researchers in Korea analyzed literature evaluating 516 AI algorithms for medical image analysis and found that only 6% validated their AI and 0% were ready for clinical use. This lack of appropriate clinical validation is referred to as digital exceptionalism. An analysis of 47 biomedical unicorns found that most of the highest valued startups in healthcare have a limited or non‐existent participation in the publicly available scientific literature.

How startups are leveraging deep tech knowledge to power ahead


Geeta Manjunath turned entrepreneur in the backdrop of a tragedy. In 2017, a cousin she was really close to succumbed to breast cancer at a relatively young age. Breast cancer is the most commonly occurring cancer in women and the second most common worldwide. Gopinath, who has a PhD in computer science from the Indian Institute of Science, applied her scientific mind to the issue. Ubiquitous screening and early detection vastly reduces fatality from cancer.

Artificial Intelligence May Hold Promise for Early Identification of Cervical Cancer in Women


Researchers from the National Institutes of Health (NIH) and Global Good have created a computer algorithm capable of identifying precancerous changes in women which place them at risk of developing cervical cancer. Known as automated visual evaluation, this new form of artificial intelligence (AI), "has the potential to revolutionize cervical cancer screening" for women in low income communities worldwide by giving their healthcare providers the ability to use digitized images collected during routine, annual screenings for cervical cancer to identify potential precancerous changes. According to America's National Cancer Institute (which is part of the NIH), this technology holds the promise of enabling physicians to more quickly catch and treat such potential changes before they develop into cancer, and could eventually replace visual inspection with acetic acid (VIA) -- the current method of screening used by healthcare professionals who work with limited resources in challenging medical care environments -- a testing system which is "known to be inaccurate." The researchers involved in this project "trained" the machine learning algorithm (automated visual evaluation) to recognize patterns in medical images and other "complex visual inputs" by digitizing and entering more than 60,000 images from an NCI archive of photographs which had been collected from more than 9,400 women in Costa Rica during a 1990s cervical cancer screening study which included follow-up studies for roughly 18 years. These images subsequently enabled the algorithm to "learn" which "cervical changes became precancers and which did not," according to NIH representatives, who added that the AI approach to cervical cancer screening was developed by NCI researchers in collaboration with the Intellectual Ventures Fund, Global Good, with findings confirmed independently by personnel from the National Library of Medicine (NLM), another component of the NIH.

Why AI will make healthcare personal


For generations healthcare has been episodic – someone gets sick or breaks a bone, they see a doctor, and then they might not see another one until the next time they get sick or injured. Now, as emerging technologies such as artificial intelligence open up new possibilities for the healthcare industry in the Fourth Industrial Revolution, policymakers and practitioners are developing new ways to deliver continuous healthcare for better outcomes. Consumers already expect access to healthcare providers to be as smart and easy as online banking, retrieving boarding passes and making restaurant reservations, according to Kaiser Permanente CEO Bernard J Tyson. Nearly three-quarters of Americans with health insurance (72%), for example, say it's important that their health insurance provider uses modern communication tools, such as instant message and two-way video. Innovative healthcare organizations such as Kaiser Permanente are listening.

Ushering in the next generation of precision trials for pediatric cancer


Cancer treatment decisions are increasingly based on the genomic profile of the patient's tumor, a strategy called "precision oncology." Over the past few years, a growing number of clinical trials and case reports have provided evidence that precision oncology is an effective approach for at least some children with cancer. Here, we review key factors influencing pediatric drug development in the era of precision oncology. We describe an emerging regulatory framework that is accelerating the pace of clinical trials in children as well as design challenges that are specific to trials that involve young cancer patients. Last, we discuss new drug development approaches for pediatric cancers whose growth relies on proteins that are difficult to target therapeutically, such as transcription factors. The landscape of genomic alterations in cancers that arise in children, adolescents, and young adults is slowly becoming clearer as a result of dedicated pediatric cancer genome-sequencing projects conducted over the past decade. Of particular note are two recent studies that produced a comprehensive picture of the genomic features that characterize many of the more common pediatric cancers (1, 2). Two major themes have emerged.

The genomic landscape of pediatric cancers: Implications for diagnosis and treatment


The past decade has witnessed a major increase in our understanding of the genetic underpinnings of childhood cancer. Genomic sequencing studies have highlighted key differences between pediatric and adult cancers. Whereas many adult cancers are characterized by a high number of somatic mutations, pediatric cancers typically have few somatic mutations but a higher prevalence of germline alterations in cancer predisposition genes. Also noteworthy is the remarkable heterogeneity in the types of genetic alterations that likely drive the growth of pediatric cancers, including copy number alterations, gene fusions, enhancer hijacking events, and chromoplexy. Because most studies have genetically profiled pediatric cancers only at diagnosis, the mechanisms underlying tumor progression, therapy resistance, and metastasis remain poorly understood. We discuss evidence that points to a need for more integrative approaches aimed at identifying driver events in pediatric cancers at both diagnosis and relapse. We also provide an overview of key aspects of germline predisposition for cancer in this age group. Approximately 300,000 children from infancy to age 14 are diagnosed with cancer worldwide every year (1). Some of the cancer types affecting the pediatric population are also seen in adolescents and young adults (AYA), but it has become increasingly clear that cancers in the latter age group have unique biological characteristics that can affect prognosis and therapy (2).

Algorithms Can Now Identify Cancerous Cells Better Than Humans


If you became concerned about a mole on your back -- perhaps it had become painful or looked unsightly -- your doctor might decide to remove it and have it evaluated. She'd send it to a pathology lab, where a sample of the tissue would be prepared in the form of a slide. It would then be sent to a pathologist, who would examine the slide to determine whether the tissue had any problematic elements, like cancer. After taking a look, the pathologist might ship it to a specialist at another lab for a second opinion. Each time the slide is moved, it is packed up and shipped to a different address.