5 barriers to AI adoption in pediatric cancer imaging
There are five key barriers that are preventing the proliferation of artificial intelligence in pediatric cancer imaging, according to a recently published editorial.
The excitement surrounding AI in adult oncology is “palpable,” experts note, with deep learning models supporting everything from lesion detection to clinical trial enrollment.
“Yet pediatric oncology remains on the periphery of this revolution as the translation of AI from adult success stories to children is obstructed by unique and persistent barriers,” Alexander J. Towbin, MD, Cincinnati Children’s, and Amit Gupta, MD, with the All India Institute of Medical Sciences, New Delhi, wrote Feb. 16 in Cancer Imaging.
The two offered five factors they believe are holding back AI in children’s cancer care:
1. Simple epidemiology: Pediatric cancers are rare, accounting for 1% of new diagnoses, and tumor types are even more uncommon. “Such scarcity limits the quantity of imaging data available for AI training and validation, creating a fundamental mismatch between the scale of pediatric oncology and the data appetite of contemporary deep learning systems.”
2. Fragmentation of available data: These cases are scattered across hundreds of specialized cancer centers, with over 200 such institutions managing the thousands of instances in the U.S. “Without systematic data sharing, models remain narrow reflections of local populations, unable to generalize across institutions or imaging platforms.”
3. Restricted data access: Public pediatric imaging datasets remain “extremely limited,” with children representing less than 1% of available information, due to several factors. “The absence of shareable, annotated datasets hinders reproducibility and prevents meaningful benchmarking across studies.”
4. Heterogeneity also a hindrance: Pediatric imaging protocols vary widely between institutions, generating a “mosaic of images” that “defies harmonization.” “Models trained in one center may fail in another purely because of technical variation rather than clinical or biological differences.”
5. Pediatric AI is not small adult AI: Kids differ in many ways from grownups, including tumor biology and disease presentation. “When adult-trained algorithms are applied to pediatric data, they often falter, missing smaller lesions or misclassifying normal developmental features as pathology.”
Towbin and Gupta closed by making a call to action for the pediatric radiology community: “embrace collaboration as its central principle.”
“The question is no longer whether AI can be implemented in pediatric cancer imaging, but how it can be implemented safely, equitably and effectively,” they concluded. “Overcoming these barriers will define the future of pediatric imaging,” the two added later. “If achieved, pediatric AI can serve as a model grounded in scientific rigor and reflective of our collective responsibility to children with cancer.”
Read the rest in the open access article here.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.