A Free Calculator for a Fear That Comes With Breast Cancer

US · CanadaAmérica Latina

Weekly AI Watch

Every week, what changed and what it means for you. No hurry, no alarm.

Week of 30 October 2026EN | ESPast editions15,230 papers found · 14,455 read · 4,301 belonged · 46 stories passed every check

Observatorio de la
Inteligencia Artificial

Cada semana, qué cambió y qué significa para usted. Sin prisa y sin alarma.

Semana del 30 de octubre de 2026EN | ESEdiciones anteriores15,230 estudios encontrados · 14,455 leídos · 4,301 correspondían · 46 notas pasaron todas las comprobaciones

weeklyAI · Week of 30 October 2026weeklyAI · Semana del 30 de octubre de 2026

← The good news← Las buenas noticias

other · Journal of translational medicine · la publicación, 24 jul 2026 · gratis

Una calculadora gratuita estima el riesgo de que el cáncer de mama llegue al cerebro

Salió de 186,351 pacientes de dos registros de Estados Unidos y nueve centros oncológicos de China. La prueba fue interna, con los mismos datos, así que su acierto en un hospital cualquiera todavía no está medido.

Versión breve · la versión detallada sigue, unos 6 min

Pregunte al chat de weeklyAI

Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.

Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.

El estudio, de un vistazo
Quiénes: pacientes con cáncer de mama primario invasivo
Cuántos: 186,351
Dónde: Estados Unidos (SEER y NCDB) y China (nueve centros oncológicos)
Cuándo: diagnosticadas entre 2010 y 2018
Tipo de estudio: analysis of what people did
Quién lo hizo: hospitales de China y registros de Estados Unidos
El límite que importa: La prueba fue interna, con los mismos datos; no se ha probado en un hospital nuevo.

La calculadora estima una probabilidad, no un plazo.
Lectura de weeklyAI

Ver · 0:29

Si alguien de su familia está en tratamiento por cáncer de mama, el miedo tiene nombre: que las células lleguen al cerebro. Un equipo de investigadores juntó los datos de 186,351 pacientes con cáncer de mama primario, de dos registros de Estados Unidos (SEER y la Base Nacional de Datos del Cáncer) y de nueve centros oncológicos de China. El grupo incluyó a mujeres con cáncer de mama invasivo diagnosticado entre 2010 y 2018. De ese total, 2,982 personas, el 1.6%, desarrollaron metástasis en el cerebro.

Los investigadores probaron nueve métodos de aprendizaje automático, la tecnología que busca patrones en muchos datos a la vez. Ganó el más sencillo, con cinco datos de la paciente: el grado de afectación de los ganglios linfáticos, si el tumor ya se había extendido a otros órganos, el tamaño del tumor, la edad y si se operó o no. El programa quedó en internet, de uso libre y sin costo, en una página del proyecto.

En una parte de los datos que el programa no había visto durante su entrenamiento, acertó a separar a quienes desarrollaron metástasis cerebral de quienes no, con una calificación de 0.907 en una escala donde 1 es el máximo y 0.5 es el azar. Esa parte, sin embargo, salió de las mismas tres fuentes. No fue probado en una población nueva, así que la cifra probablemente sea algo optimista.

Los propios autores piden cautela. Dicen que la herramienta no debe usarse por sí sola como motivo para hacer una resonancia magnética del cerebro, y que las imágenes cerebrales de rutina no se recomiendan en general para pacientes sin síntomas. Los cortes de riesgo que proponen (5%, 10% y 20%) son exploratorios y todavía no se validan. Además, el estudio contó la metástasis como un sí o un no, sin tomar en cuenta que unas pacientes fueron seguidas más tiempo que otras: 55.1 meses de mediana en el grupo con metástasis cerebral y 65.4 meses en el resto. Quien es observado más tiempo tiene más oportunidades de que se le cuente. La herramienta estima una probabilidad, no una fecha.

Busque la calculadora gratuita y pregunte en su centro de cáncer si alguien la ha probado con sus propias pacientes antes de tomar su número como un plan.

Qué significa para usted

La calculadora gratuita puede darle un número, y ese número no es un plan ni una fecha. Pregunte en su centro de cáncer si el equipo la ha probado con sus propias pacientes y qué opina de usarla antes de que alguien la tome como motivo para pedir estudios. Lo que se midió fue el acierto del programa en los mismos datos con que se armó.

Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0

Quién pagó: El estudio fue financiado por la Fundación de Ciencias Naturales de la Provincia de Fujian de China (No. 2023J011345); el artículo no indica si el financiador tuvo algún papel y no menciona proveedores de equipos o programas informáticos.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1192 palabras · unos 6 minLeerla →Cerrar

Una calculadora gratuita estima el riesgo de que el cáncer de mama llegue al cerebro

Se probó en 186,351 pacientes de China y Estados Unidos. Predice probabilidades, no tiempos, y todavía no se ha probado en un hospital nuevo.

Un número que hoy no existe en la consulta: la probabilidad de que un cáncer de mama se extienda al cerebro. Un grupo de investigadores de hospitales de China y de dos registros de Estados Unidos construyó una calculadora en línea que da esa cifra en segundos. El estudio se publicó el 24 de julio de 2026 en el Journal of Translational Medicine.

El cerebro es un refugio difícil. Ahí los tratamientos llegan peor, y cuando el cáncer se instala aparecen dolores de cabeza fuertes, convulsiones y pérdida de memoria y de movimiento, según describen los propios autores¹. Los autores citan estimaciones de la literatura según las cuales entre 10 y 30 por ciento de las pacientes con enfermedad avanzada desarrollan metástasis cerebral, y hasta 50 por ciento en subtipos agresivos como el HER2 positivo y el triple negativo². La mediana de supervivencia tras ese diagnóstico suele medirse en meses³.

La herramienta no es un aparato ni un estudio de imagen. Es una página web gratuita que se abre en el navegador⁴. El médico escribe cinco datos de la paciente: la edad, el estadio del tumor (T), el de los ganglios (N), el de la metástasis (M) y si se operó⁵⁴. El programa devuelve un porcentaje de riesgo y, además, un dibujo que muestra cuánto empuja cada dato hacia arriba o hacia abajo⁶. Lo curioso es que no hace falta un modelo complicado: de nueve sistemas que los autores probaron, el que mejor funcionó fue el más simple, una regresión logística con un filtro que descarta variables, llamado LASSO⁷.

El estudio juntó 186,351 pacientes con cáncer de mama invasivo diagnosticadas entre 2010 y 2018, de nueve centros oncológicos chinos y de los registros SEER y NCDB de Estados Unidos⁸. De ellas, 2,982 (1.6 por ciento) desarrollaron metástasis cerebral. Los datos se dividieron al azar: 70 por ciento para enseñar al programa y 30 por ciento para examinarlo. En ese examen, la calculadora acertó a separar a quienes desarrollaron metástasis de quienes no con una medida de 0,907, en una escala donde 0,5 es adivinar al azar y 1 sería perfecto⁷. Los autores reportan que el modelo estuvo bien calibrado y que ofreció beneficio clínico neto en el análisis de decisiones⁹.

Aquí conviene detenerse. Esa prueba se hizo con pacientes de las mismas tres bases de datos, no con un hospital nuevo¹⁰. Los propios autores lo llaman validación interna y advierten que no es validación externa independiente¹⁰. Por eso la cifra de 0,907 probablemente sea algo optimista: el programa ya había visto casos parecidos. Los autores presentan los análisis de supervivencia solo como contexto, para mostrar lo grave que es la metástasis cerebral, y no como un objetivo principal del estudio.

Hay otra consecuencia de cómo se armó el estudio. La metástasis cerebral se contó como sí o no, sin tomar en cuenta cuánto tiempo estuvo cada paciente en observación¹¹. Las pacientes con metástasis tuvieron un seguimiento mediano de 55,1 meses y las demás de 65,4¹¹. Quien es vigilada más tiempo tiene más oportunidades de que le encuentren algo. La calculadora estima una probabilidad, no un plazo. Y como los datos vienen de expedientes ya escritos, los autores aceptan que puede haber factores no medidos y errores de captura¹².

Sobre el uso clínico, los autores son cuidadosos. Las guías no recomiendan tomar imágenes del cerebro a pacientes sin síntomas, y el modelo no debe usarse como razón automática para pedir una resonancia¹³. Puede servir, dicen, para ordenar el riesgo y decidir la vigilancia caso por caso¹³. Proponen umbrales exploratorios: a partir de 10 por ciento considerar vigilancia reforzada, sobre todo en pacientes jóvenes, con ganglios avanzados, metástasis fuera del cerebro, HER2 positivo o triple negativo¹⁴. Y aclaran que esos cortes tendrían que validarse en estudios prospectivos antes de adoptarse como regla¹⁵.

Para dimensionar la herramienta, ayuda mirar otros trabajos con inteligencia artificial en cáncer de mama. Un resumen de una revisión sobre lectura automática de mastografías, del que solo pudimos leer el resumen porque el texto completo está detrás de una suscripción, reporta que la inteligencia artificial pasó en una década de prueba de concepto a evaluación aleatorizada¹⁶. Esa revisión junta 14 estudios y 1.214.885 exámenes, con un promedio de 0,890 en la misma escala de discriminación¹⁷. Otro resumen, también leído solo en su resumen, describe un dispositivo de reconocimiento de mama por ultrasonido probado en 6,817 personas en 107 hospitales¹⁸¹⁹. Los dos casos comparten una advertencia: el desempeño promedio no se traslada solo a un programa nuevo, y hace falta validarlo localmente²⁰²¹. La calculadora de metástasis cerebral está en esa misma familia: promete mucho en papel y aún no ha pasado por un hospital que no sea de los suyos.

Así lo leemos nosotros. La escena que hay que imaginar es la de una mujer joven, con ganglios positivos, sentada frente a un médico que teclea cinco datos en una página gratuita. El porcentaje aparece y, con él, una pregunta: ¿esto cambia algo? La herramienta no es un regalo desinteresado ni una estafa: es investigación académica publicada, financiada por la Fundación de Ciencias Naturales de la provincia de Fujian, en China, y los autores declaran no tener conflictos de interés²². Pero una calculadora gratuita en línea también recibe datos. Antes de escribir el nombre o la edad de una familiar, vale revisar qué campos son obligatorios y si hay aviso de privacidad. Pregunte en la consulta quién guarda esa información y para qué. Si no hay política clara, use datos mínimos o pida que el médico la llene.

Es razonable esperar que la use sobre todo personal con práctica digital, en hospitales grandes, y que en clínicas pequeñas o rurales casi nadie la conozca. Si en su clínica no la conocen, pida que la revisen junto con el equipo tratante. Y trate el porcentaje como una pista, no como sentencia: pida que le expliquen qué factores lo suben y cuáles lo bajan, y cómo se compara con su historia real.

Queda lo que falta. Los autores piden validación prospectiva, es decir, probar la herramienta en pacientes nuevas y seguir su curso¹⁰. También señalan que una validación que deje fuera una de las fuentes queda como trabajo futuro prioritario. También señalan que no tuvieron detalle de los tratamientos dirigidos contra HER2, lo que pudo influir en que el subtipo molecular no quedara entre las cinco variables finales. Por ahora, la calculadora está abierta al público en la dirección que da el artículo⁴, y los autores aclaran que los umbrales que proponen son exploratorios y tendrían que validarse en estudios prospectivos¹⁵.

Vuelva a la escena de la consulta, ahora con lo que sabe. La página muestra un número y un dibujo; el número no tiene fecha, solo probabilidad. La pregunta que puede llevar a su próxima cita no es "¿qué porcentaje me dio?", sino: "¿Este cálculo se probó en pacientes como yo, en un hospital como el mío, y qué haríamos distinto si sale alto?"

De dónde sale cada dato de contexto, y cuánto leímos de cada documento

Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Beyond its impact on survival, the clinical burden of BM is immense, inflicting a constellation of debilitating neurological symptoms—from severe headaches and seizures to profound cognitive impairment and focal motor deficits—that catastrophically erode a patient’s quality of life and functional independence."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The incidence of brain metastasis (BM) in BC patients is alarmingly high, affecting an estimated 10–30% of those with advanced disease and rising to as high as 50% in aggressive molecular subtypes such as HER2-positive and triple-negative breast cancer (TNBC)."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The onset of BM marks a critical turning point in a patient’s illness, invariably leading to a dire prognosis with a median overall survival often measured in mere months."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The tool, publicly available at https://hb-research.shinyapps.io/Brain_metastasis_app/ , allows clinicians to input a patient’s key predictors (AJCC T/N/M stage, age, and surgery status) to receive an instantaneous probability of BM (Fig. 9 )."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Key predictors retained in the final parsimonious model included AJCC N/M/T stage, age, and surgery status."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A key feature of the application is the real-time generation of a patient-specific SHAP waterfall plot, which provides a transparent visualization of the factors driving the risk prediction."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The LASSO-regularized Logistic Regression model achieved superior discrimination in the validation set, with an AUC of 0.907 (95% CI: 0.857–0.958)."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The final cohort comprised 186,351 patients with primary BC."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The model demonstrated good calibration, clinical net benefit, and interpretability, with SHAP analysis confirming the contribution of each retained predictor to individual risk estimates."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This constitutes an internal hold-out validation rather than independent external validation."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "BM was modeled as a binary outcome rather than a time-to-event endpoint. This approach does not account for censoring, competing risks, or differences in follow-up duration between the BM and non-BM groups (median 55.1 vs. 65.4 months)."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The retrospective nature of the data sources introduces the potential for unmeasured confounding variables and coding inaccuracies."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Because routine brain imaging is not generally recommended for asymptomatic BC patients, the present model should not be used as an automatic indication for MRI screening. Instead, it may serve as a risk-stratification aid to support individualized surveillance decisions."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Patients with predicted risk ≥ 10% may be considered for enhanced surveillance, including individualized brain MRI assessment, particularly among younger patients or those with advanced nodal disease, extracranial metastasis, HER2-positive disease, or triple-negative breast cancer."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "These thresholds are exploratory and should be prospectively validated before being adopted as definitive clinical decision cutoffs."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Deep-learning artificial intelligence (AI) for mammographic screening moved from proof of concept to randomised evaluation in a single decade."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Pooled standalone AUC across 14 studies and 1,214,885 examinations was 0.890 (95% CI 0.858-0.915), with I 2 = 97.4% and a 95% prediction interval of 0.731-0.960."
Zhou J, Si P, Zhang Y, Song J, He T, Liu Q, et al. (2026). A non-invasive end-to-end intelligent assistance system for breast ultrasound. Nature Communications. 10.1038/s41467-026-73170-5 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Here, we develop and test an end-to-end breast intelligent recognition device (BIRD) for women across diverse institution/population datasets."
Zhou J, Si P, Zhang Y, Song J, He T, Liu Q, et al. (2026). A non-invasive end-to-end intelligent assistance system for breast ultrasound. Nature Communications. 10.1038/s41467-026-73170-5 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "BIRD is applied in breast cancer screening for 6,817 individuals and shows high consistency (Cohen's kappa: 0.702 (95% confidence interval: 0.628-0.777)) with clinical assessments in real-world application across 107 hospitals."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Pooled accuracy is high on average but so dispersed that it cannot be transferred to a new programme; local validation before deployment remains necessary."
Zhou J, Si P, Zhang Y, Song J, He T, Liu Q, et al. (2026). A non-invasive end-to-end intelligent assistance system for breast ultrasound. Nature Communications. 10.1038/s41467-026-73170-5 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Although artificial intelligence enhances medical image classification to effectively improve lesion diagnosis accuracy and efficiency, it still faces generalization challenges in breast ultrasound and its clinical potential remains underutilized in complex real-world scenarios."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Funding This study was supported by the Natural Science Foundation of Fujian Province of China (No. 2023J011345)."

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.

other · Journal of translational medicine · the paper, 24 Jul 2026 · free

A Free Calculator for a Fear That Comes With Breast Cancer

A model built from 186,351 patients flagged brain spread with an accuracy score of 0.907 on a slice of the same records. No new hospital has tried it yet.

Short version · the longer version follows, about 6 min

Ask the weeklyAI chat

Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.

Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.

The study at a glance
Who: women with primary breast cancer
How many: 186,351
Where: United States and China
When: diagnosed between 2010 and 2018
Kind of study: analysis of what people did
Who did it: Harbin Medical University Cancer Hospital and other centers in China
The limit that matters: Tested only on a slice of the same records; no new hospital has tried it.

The tool estimates a chance, not a date.
weeklyAI's reading

Watch · 0:28

You or someone you know is being treated for breast cancer. Among the worries that follow the diagnosis, one is that the cancer reaches the brain.

Researchers gathered records from 186,351 women with primary breast cancer, diagnosed between 2010 and 2018. The records came from two United States registries, SEER and the National Cancer Database, and from nine cancer centers in China. Of those women, 1.6 percent, about 2,982, developed brain metastasis.

The researchers tested nine machine-learning methods side by side. The winner was the simplest: a model that uses five pieces of information a doctor already has. They are the lymph node stage, the distant-metastasis stage, the tumor stage, the patient's age, and whether surgery was done.

On a held-out 30 percent of the same records, this model scored 0.907 on a standard measure of how well it separates patients who developed brain spread from those who did not. A perfect score is 1.0. The calculator is online and free to use.

That held-out slice came from the same three databases the model was built on. No separate hospital or country has tested it. The authors say the figure may be somewhat optimistic. They also say the tool needs testing in new patients before it is used in routine care.

The authors are direct: the tool should not be used as an automatic reason for a brain MRI. Routine brain imaging is not generally recommended for patients without symptoms. The risk cutoffs they suggest, at 5, 10 and 20 percent, are described as exploratory and not yet validated.

The tool estimates a chance, not a date. Brain spread was counted as a yes or no over unequal follow-up: a median of 55.1 months in the group that developed it and 65.4 months in the others. Women watched longer had more chances to be counted.

The researchers put the calculator online for anyone to use. Then ask whether your own cancer center has tested it on its own patients before anyone treats its number as a plan.

What this means for you

The calculator asks for five things a doctor already knows: the node stage, the distant-spread stage, the tumor stage, age, and whether surgery was done. Its score comes from the same records it learned on, and no other hospital has tried it, so treat the number as a question to bring to the cancer team rather than a plan. Ask whether your own center has tested it on its own patients.

Who paid: The study was supported by the Natural Science Foundation of Fujian Province of China (No. 2023J011345); the article does not say whether the funder had any role, and it names no equipment or software lenders.

Do not take this as professional medical advice.

The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1215 words · about 6 minRead it →Close

A Free Calculator Estimates a Breast Cancer Patient's Risk of Brain Spread. Here's What It Can't Tell Her.

A model built from 186,351 patient records was good at sorting people by risk. It has never been tested on a new population, and its own authors say it is not a reason to order a scan.

The number 186,351 is the patient count behind a new model for estimating whether breast cancer will spread to the brain. Researchers pooled records from two United States cancer registries and nine cancer centers in China, covering women diagnosed with invasive breast cancer between 2010 and 2018. Of those 186,351 patients, 2,982, or 1.6 percent, developed brain metastasis¹.

Brain metastasis is the spread of breast cancer cells to the brain, and it changes a patient's course sharply. In advanced disease an estimated 10 to 30 percent of patients are affected, and in aggressive subtypes such as HER2-positive and triple-negative breast cancer the figure rises as high as 50 percent². Once it appears, median overall survival is often measured in months³. It also brings severe headaches, seizures, cognitive impairment and focal motor deficits that erode a patient's independence⁴. Any tool that flags risk early would matter to a family.

The model keeps five inputs: the cancer's N, M and T stages, the patient's age, and whether surgery was performed⁵. The researchers trained nine machine learning algorithms on 70 percent of the records and tested them on the remaining 30 percent. The best performer was a LASSO-regularized logistic regression, a linear method that discards weak variables, with a validation score of 0.907 (95% CI: 0.857–0.958) on the area-under-the-curve measure⁶. It also showed good calibration and clinical net benefit, and an explainability analysis confirmed how each of the five inputs pushed individual estimates⁷.

The model was turned into a free web calculator. The calculator gives an instant probability of brain metastasis from the five inputs⁸. The authors report that the tool also shows which of the five factors raised or lowered that number⁹. The inputs are ordinary facts from a pathology report and an operative note, the kind already in the chart.

The limits are stated by the authors themselves. The 30 percent test set came from the same three databases as the training set, which the paper calls internal hold-out validation rather than independent external validation¹⁰. The retrospective records carry potential for unmeasured confounding and coding inaccuracies¹¹. Brain metastasis was counted as a yes-or-no event over unequal observation times, with median follow-up of 55.1 months in the brain-metastasis group versus 65.4 months in the others¹².

Consider what that unequal follow-up means at the kitchen table. A woman watched for 65 months had more appointments, more scans and more chances to be counted than one watched for 55. The tool estimates a chance, not a date. And because the same hospitals supplied both halves of the data, the 0.907 figure is likely somewhat generous compared with what a new hospital would see.

The authors are explicit about use. Routine brain imaging is not generally recommended for patients without symptoms, and the model should not be an automatic reason to order an MRI¹³. They suggest that patients above a 10 percent predicted risk might be considered for enhanced surveillance, especially younger patients or those with advanced nodal disease, extracranial spread, HER2-positive or triple-negative cancer¹⁴. They then call those cutoffs exploratory, requiring prospective validation before adoption¹⁵.

The money behind the work is public: the Natural Science Foundation of Fujian Province of China, grant number 2023J011345¹⁶. The authors declare no competing interests, and the paper does not say who has run the calculator in practice. The study appeared in the Journal of Translational Medicine, which lists it as received in July 2026 and accepted in October 2026.

This belongs to a family of studies asking whether software can read breast tissue or breast records better than routine practice. A review of ten years of deep-learning mammography screening, which we could read only in its summary, found that the technology moved from proof of concept to randomized evaluation within a decade¹⁷. Its pooled standalone score across 14 studies and 1,214,885 examinations was 0.890¹⁸. In randomized and paired prospective trials, the pooled detection-rate ratio was 1.13 (0.83–1.55)¹⁹. The same summary cautions that pooled accuracy is so dispersed it cannot be transferred to a new program without local validation²⁰, and that the larger gains came from non-randomized designs, the pattern confounding would produce²¹. Its conclusion: AI works as a second reader and triage tool, not a replacement for the radiologist²².

A second effort, a non-invasive breast recognition device tested across 107 hospitals, which we could also read only in its summary, reached 0.837 accuracy on its internal test and improved radiologists' accuracy in two reader studies²³. It was applied to screening for 6,817 individuals with high consistency against clinical assessment²⁴. Its authors still name generalization as the open challenge in real-world settings²⁵. The pattern across these three: a good score inside the data, and an unanswered question outside it.

Here is how we read it. A calculator is not a doctor and not a verdict, yet a percentage arriving from a computer tends to carry more authority than it has earned. We expect that a family typing a mother's or sister's details into this page may treat the result as a conclusion rather than one voice among several: pressing for scans the care team would not have proposed, or relaxing when the number comes back low. You would know we are wrong if patients and clinicians keep treating the output as a single question to raise, ask what it cannot see, and change nothing on the strength of the number alone.

We also expect a subtler trap. A high score may lead to more frequent brain scans, which find more small metastases, which makes the warning look accurate; a low score may mean fewer scans and a later diagnosis the tool is never blamed for. The way to tell is whether two patients with different scores are followed on the same schedule and the high scorer still turns out to have more spread. Until then, ask your care team whether the surveillance plan would differ if the calculator did not exist, and how its accuracy was checked in people like you, in your country.

We also read the wider record on this kind of software, and it says the same thing twice: a good average score does not travel. A tool validated where it was built is a promise, not a proof, and the honest test is one run in the hospital where you are treated, on patients like you.

What can you do with a percentage? More than sit with it. Use the estimate to write two or three practical questions for the next visit: which symptoms should prompt a call, what follow-up is already planned, and who arranges travel if extra scans are suggested. Keep a copy of the answers. The number is one input; the plan is the point.

The calculator sits behind a web address, free to open, waiting for five facts from a chart. What it returns is a chance, not a sentence, and the question to carry into the clinic is simple: what would you do differently because of this number, and why?

Where each piece of context comes from, and how much of it we read

Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The final cohort comprised 186,351 patients with primary BC."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The incidence of brain metastasis (BM) in BC patients is alarmingly high, affecting an estimated 10–30% of those with advanced disease and rising to as high as 50% in aggressive molecular subtypes such as HER2-positive and triple-negative breast cancer (TNBC)."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The onset of BM marks a critical turning point in a patient’s illness, invariably leading to a dire prognosis with a median overall survival often measured in mere months."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "Beyond its impact on survival, the clinical burden of BM is immense, inflicting a constellation of debilitating neurological symptoms—from severe headaches and seizures to profound cognitive impairment and focal motor deficits—that catastrophically erode a patient’s quality of life and functional independence."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "Key predictors retained in the final parsimonious model included AJCC N/M/T stage, age, and surgery status."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The LASSO-regularized Logistic Regression model achieved superior discrimination in the validation set, with an AUC of 0.907 (95% CI: 0.857–0.958)."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The model demonstrated good calibration, clinical net benefit, and interpretability, with SHAP analysis confirming the contribution of each retained predictor to individual risk estimates."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The tool, publicly available at https://hb-research.shinyapps.io/Brain_metastasis_app/ , allows clinicians to input a patient’s key predictors (AJCC T/N/M stage, age, and surgery status) to receive an instantaneous probability of BM (Fig. 9 )."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "A key feature of the application is the real-time generation of a patient-specific SHAP waterfall plot, which provides a transparent visualization of the factors driving the risk prediction."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "This constitutes an internal hold-out validation rather than independent external validation."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "The retrospective nature of the data sources introduces the potential for unmeasured confounding variables and coding inaccuracies."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "BM was modeled as a binary outcome rather than a time-to-event endpoint. This approach does not account for censoring, competing risks, or differences in follow-up duration between the BM and non-BM groups (median 55.1 vs. 65.4 months)."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "Because routine brain imaging is not generally recommended for asymptomatic BC patients, the present model should not be used as an automatic indication for MRI screening. Instead, it may serve as a risk-stratification aid to support individualized surveillance decisions."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "Patients with predicted risk ≥ 10% may be considered for enhanced surveillance, including individualized brain MRI assessment, particularly among younger patients or those with advanced nodal disease, extracranial metastasis, HER2-positive disease, or triple-negative breast cancer."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "These thresholds are exploratory and should be prospectively validated before being adopted as definitive clinical decision cutoffs."
Yuan, Q., Sha, Y., Yu, R. et al. (2026). Interpretable machine learning model for brain metastasis in breast cancer: a large-scale, multi-center study. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08675-0 - the article this story is about — the whole article — the passage: "Funding This study was supported by the Natural Science Foundation of Fujian Province of China (No. 2023J011345)."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — only the abstract - the full text could not be fetched — the passage: "Deep-learning artificial intelligence (AI) for mammographic screening moved from proof of concept to randomised evaluation in a single decade."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — only the abstract - the full text could not be fetched — the passage: "Pooled standalone AUC across 14 studies and 1,214,885 examinations was 0.890 (95% CI 0.858-0.915), with I 2 = 97.4% and a 95% prediction interval of 0.731-0.960."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — only the abstract - the full text could not be fetched — the passage: "Across randomised and paired prospective trials (k = 3), the pooled detection rate ratio was 1.13 (0.83-1.55), with the MASAI randomised trial alone reporting 1.29 (1.09-1.51) and a 44% reduction in screen reading."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — only the abstract - the full text could not be fetched — the passage: "Pooled accuracy is high on average but so dispersed that it cannot be transferred to a new programme; local validation before deployment remains necessary."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — only the abstract - the full text could not be fetched — the passage: "The larger and more precise detection gains come from non-randomised designs, which is the pattern confounding would produce, so the randomised evidence remains the anchor."
Ciurescu S, Buciu V, Ilaș DG, Pârvănescu R, Șerban D. (2026). Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015–2025). Diagnostics. 10.3390/diagnostics16183045 — only the abstract - the full text could not be fetched — the passage: "A decade of evidence supports AI as a second reader and triage tool in organised screening, not as an autonomous replacement for the radiologist."
Zhou J, Si P, Zhang Y, Song J, He T, Liu Q, et al. (2026). A non-invasive end-to-end intelligent assistance system for breast ultrasound. Nature Communications. 10.1038/s41467-026-73170-5 — only the abstract - the full text could not be fetched — the passage: "The accuracy of BIRD in the internal test set is 0.837 (95% confidence interval: 0.827-0.846), which significantly improves radiologists' accuracy in 2 reader studies (P < 0.05)."
Zhou J, Si P, Zhang Y, Song J, He T, Liu Q, et al. (2026). A non-invasive end-to-end intelligent assistance system for breast ultrasound. Nature Communications. 10.1038/s41467-026-73170-5 — only the abstract - the full text could not be fetched — the passage: "BIRD is applied in breast cancer screening for 6,817 individuals and shows high consistency (Cohen's kappa: 0.702 (95% confidence interval: 0.628-0.777)) with clinical assessments in real-world application across 107 hospitals."
Zhou J, Si P, Zhang Y, Song J, He T, Liu Q, et al. (2026). A non-invasive end-to-end intelligent assistance system for breast ultrasound. Nature Communications. 10.1038/s41467-026-73170-5 — only the abstract - the full text could not be fetched — the passage: "Although artificial intelligence enhances medical image classification to effectively improve lesion diagnosis accuracy and efficiency, it still faces generalization challenges in breast ultrasound and its clinical potential remains underutilized in complex real-world scenarios."

Do not take this as professional medical advice.

The free weekly drip by email

One email a week: the free story of every section, with its picture, and the headlines of the rest.

No payment, nothing to cancel. Unsubscribe from any issue.

El goteo semanal gratis por correo

Un correo por semana: la nota gratuita de cada sección, con su imagen, y los titulares de las demás.

Sin pago y sin nada que cancelar. Puede darse de baja desde cualquier envío.