New AI Tool Boosts Precision Medicine in Treating
Post# of 392
(Total Views: 37)
Posted On: 10/13/2025 4:51:23 PM
New AI Tool Boosts Precision Medicine in Treating CNS Cancers
Harvard Medical School researchers have developed an AI system that distinguishes glioblastoma from other brain tumors during surgery. The tool called PICTURE achieved 99.8% accuracy in differentiating glioblastoma from primary central nervous system lymphoma. This rare malignancy is frequently misdiagnosed as glioblastoma, yet the AI system outperformed nine human neuropathologists who misclassified lymphoma as glioblastoma in 38% of test cases.
With over 100 distinct subtypes, central nervous system tumors involve exceptional diagnostic complexity. Glioblastoma represents the most common variety, with patients typically receiving surgical tumor removal but facing a median survival of just eight months. Primary CNS lymphoma shares overlapping features with glioblastoma but requires different treatment. Median survival reaches three years with appropriate therapy.
When lymphoma is discovered during surgeries meant for tumor removal, surgeons typically halt further intervention to preserve neurological function. Patients are then referred for radiotherapy combined with chemotherapy instead. Dr. Kun-Hsing Yu, associate professor at Harvard and the study’s senior author, emphasized that distinguishing these tumors challenges even seasoned experts.
The research team developed PICTURE using 2,141 tissue slides collected from five medical centers worldwide. The system captures the wide spectrum of tumor characteristics from individual datasets and patterns reported in medical literature. PICTURE’s unique design allows it to recognize when cases fall outside its training parameters. Unlike conventional AI models that produce answers even when there is little supporting evidence, PICTURE can acknowledge knowledge gaps.
This capability proved crucial when the system identified 67 types of rare cancers neither seen during training nor fitting glioblastoma or lymphoma patterns. Testing across five independent groups with different tissue preparation methods demonstrated consistent accuracy. The system significantly exceeded current computational pathology methods available for clinical use. Human neuropathologists with varying clinical experience struggled with the same diagnostic challenges.
Among 113 misdiagnosed instances across evaluators, 72% were associated with low to moderate confidence levels. Several participants noted in feedback that the evaluation proved more challenging than expected, highlighting the complexity of brain tumor assessment. Standard machine learning methods train to classify outcomes into predefined categories. PICTURE’s approach flags cases likely representing neither glioblastoma nor lymphoma.
This enables clinicians to identify rare or atypical presentations warranting further evaluation, enhancing diagnostic trustworthiness while helping doctors accurately identify common cancers while remaining vigilant for rare diseases. Yu acknowledged two major hurdles that need to be addressed before widespread adoption. Regulatory approval remains necessary before PICTURE can serve as a primary clinical tool, and the system must then integrate directly into pathology workflows to deliver results seamlessly without disrupting routine practice.
Ultimately, the tool could shift diagnosis from manual interpretation toward systematic, data-driven processes, enabling faster accurate diagnosis for similar-looking tumors that need different treatments and providing real-time surgical guidance.
As such innovations improve brain cancer diagnosis, therapeutics developed by companies like CNS Pharmaceuticals Inc. (NASDAQ: CNSP) could stand a higher chance of delivering the needed clinical outcomes to patients afflicted by these malignancies.
NOTE TO INVESTORS: The latest news and updates relating to CNS Pharmaceuticals Inc. (NASDAQ: CNSP) are available in the company’s newsroom at https://ibn.fm/CNSP
Please see full terms of use and disclaimers on the TinyGems website applicable to all content provided by TinyGems, wherever published or re-published: https://www.TinyGems.net/Disclaimer
Harvard Medical School researchers have developed an AI system that distinguishes glioblastoma from other brain tumors during surgery. The tool called PICTURE achieved 99.8% accuracy in differentiating glioblastoma from primary central nervous system lymphoma. This rare malignancy is frequently misdiagnosed as glioblastoma, yet the AI system outperformed nine human neuropathologists who misclassified lymphoma as glioblastoma in 38% of test cases.
With over 100 distinct subtypes, central nervous system tumors involve exceptional diagnostic complexity. Glioblastoma represents the most common variety, with patients typically receiving surgical tumor removal but facing a median survival of just eight months. Primary CNS lymphoma shares overlapping features with glioblastoma but requires different treatment. Median survival reaches three years with appropriate therapy.
When lymphoma is discovered during surgeries meant for tumor removal, surgeons typically halt further intervention to preserve neurological function. Patients are then referred for radiotherapy combined with chemotherapy instead. Dr. Kun-Hsing Yu, associate professor at Harvard and the study’s senior author, emphasized that distinguishing these tumors challenges even seasoned experts.
The research team developed PICTURE using 2,141 tissue slides collected from five medical centers worldwide. The system captures the wide spectrum of tumor characteristics from individual datasets and patterns reported in medical literature. PICTURE’s unique design allows it to recognize when cases fall outside its training parameters. Unlike conventional AI models that produce answers even when there is little supporting evidence, PICTURE can acknowledge knowledge gaps.
This capability proved crucial when the system identified 67 types of rare cancers neither seen during training nor fitting glioblastoma or lymphoma patterns. Testing across five independent groups with different tissue preparation methods demonstrated consistent accuracy. The system significantly exceeded current computational pathology methods available for clinical use. Human neuropathologists with varying clinical experience struggled with the same diagnostic challenges.
Among 113 misdiagnosed instances across evaluators, 72% were associated with low to moderate confidence levels. Several participants noted in feedback that the evaluation proved more challenging than expected, highlighting the complexity of brain tumor assessment. Standard machine learning methods train to classify outcomes into predefined categories. PICTURE’s approach flags cases likely representing neither glioblastoma nor lymphoma.
This enables clinicians to identify rare or atypical presentations warranting further evaluation, enhancing diagnostic trustworthiness while helping doctors accurately identify common cancers while remaining vigilant for rare diseases. Yu acknowledged two major hurdles that need to be addressed before widespread adoption. Regulatory approval remains necessary before PICTURE can serve as a primary clinical tool, and the system must then integrate directly into pathology workflows to deliver results seamlessly without disrupting routine practice.
Ultimately, the tool could shift diagnosis from manual interpretation toward systematic, data-driven processes, enabling faster accurate diagnosis for similar-looking tumors that need different treatments and providing real-time surgical guidance.
As such innovations improve brain cancer diagnosis, therapeutics developed by companies like CNS Pharmaceuticals Inc. (NASDAQ: CNSP) could stand a higher chance of delivering the needed clinical outcomes to patients afflicted by these malignancies.
NOTE TO INVESTORS: The latest news and updates relating to CNS Pharmaceuticals Inc. (NASDAQ: CNSP) are available in the company’s newsroom at https://ibn.fm/CNSP
Please see full terms of use and disclaimers on the TinyGems website applicable to all content provided by TinyGems, wherever published or re-published: https://www.TinyGems.net/Disclaimer
(0)
(0)