AI sees race. Cancer diagnostic algorithms were found to have bias, performing unevenly based on patient demographics.


 A disturbing study from Harvard Medical School has uncovered a "hidden" bias in AI models used for cancer diagnosis. The research found that deep learning algorithms, when trained on medical images like pathology slides, can learn to identify a patient's self-reported race—a feat human doctors cannot do from images alone. The problem arises when the AI uses this racial data as a "shortcut" to make diagnostic predictions, rather than relying solely on biological disease markers.

The study revealed that in nearly 30% of the tested tasks, the AI models exhibited significant performance disparities, often yielding less accurate results for Black patients due to imbalances in the training data. This "algorithmic racism" could lead to misdiagnoses and unequal care if left unchecked. The researchers are calling for a new training approach, proposing a method called "FAIR-Path" that explicitly prevents models from relying on demographic shortcuts, ensuring that AI tools remain colorblind and clinically objective.

Read the original article at: https://futurism.com/health-medicine/ai-cancer-diagnostic-bias

 

Follow us on Instagram, Twitter, and Facebook to stay up to date with what's new in healthcare all around the world.

Comments

Post a Comment

Popular posts from this blog

Guesswork gone. A new scoring system (ST-RADS) predicts soft-tissue tumor malignancy with 99.2% accuracy.

Image
 Radiologists may soon have a powerful new standard for evaluating soft-tissue tumors. A study detailed in Radiology Business introduces "ST-RADS" (Soft-Tissue Tumor Reporting and Data System), a structured MRI scoring framework designed to replace the often vague descriptive reports currently in use. In a validation study involving roughly 200 patients, the ST-RADS system demonstrated exceptional precision, achieving a 99.2% accuracy rate in predicting malignancy—significantly outperforming the 92.8% accuracy of standard radiological reports. Crucially, the system was perfect (100% accuracy) in identifying benign tumors, a capability that could drastically reduce unnecessary biopsies and patient anxiety. By standardizing how these complex images are interpreted, ST-RADS offers a clear, objective roadmap for clinicians, ensuring that aggressive cancers are flagged immediately while harmless lumps are safely monitored without invasive intervention. Read the original artic...
Read more

The system is too slow. New data reveals that missed "waiting time" targets are hiding the true, deadly cost of cancer care.

Image
 A critical new analysis argues that current targets for cancer treatment waiting times are failing patients and obscuring the true extent of delays. The report highlights that arbitrary benchmarks such as the 62 day target from referral to treatment are frequently missed and do not account for the increasing complexity of modern cancer care. Patients requiring multiple diagnostic tests often face much longer waits that are not accurately reflected in official data. These delays are not merely administrative nuisances they significantly increase the risk of mortality.  The authors contend that simply setting stricter targets is ineffective without addressing the underlying lack of resources. Instead they call for a national learning system driven by data and collaboration to identify bottlenecks in real time and prioritize patients based on clinical urgency rather than outdated metrics. Read the original article at: http://www.bmj.com/content/392/bmj.s14.short?rss=1 Follow ...
Read more

Digital Oncology Insights: 4th December - 10th December

Image
  1. CTCeptor technology simultaneously isolates tumor cells and fibroblasts for better diagnosis. Diagnosing cancer accurately is often difficult because standard liquid biopsies can miss important cells hiding in the blood. To solve this, researchers at DGIST have developed "CTCeptor," a new technology that captures both circulating tumor cells and the "helper" cells (fibroblasts) that support cancer growth. Most current tests try to filter cells based on their size or specific markers, but they often fail to catch cancer cells that look slightly different. CTCeptor uses a unique extraction method that successfully isolates these cells regardless of their size, providing a much more complete picture of the disease. The impact of this technology on patient care could be profound. In recent trials with breast cancer patients, CTCeptor detected 15 times more tumor cells than existing methods. By analyzing both the cancer cells and their environment, doctors can b...
Read more