other · Diagnostics (Basel, Switzerland) · la publicación, 3 ago 2026 · gratis
Un estudio observó que más pacientes con enfermedad pulmonar tomaron medicamentos tras usar una herramienta de inteligencia artificial
El estudio observó a 209 pacientes y no puede probar que la herramienta causara los cambios; además, faltaron datos de seguimiento en muchos casos.
Versión breve · la versión detallada sigue, unos 7 min
Pregunte al chat de weeklyAI
Pregúnteme por este estudio: a quiénes se estudió, qué encontró y qué no dice.
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 a quienes les estaban investigando una enfermedad pulmonar intersticial
- Cuántos
- 209 pacientes
- Dónde
- Estados Unidos
- Cuándo
- entre julio de 2024 y junio de 2026
- Tipo de estudio
- análisis de lo que hicieron los médicos en la práctica real
- Quién lo hizo
- Investigadores de universidades de Estados Unidos (Northwestern, Indiana, Jefferson, Harvard)
- El límite que importa
- Faltaron datos de seguimiento en muchos casos y el estudio no prueba causa
Porcentaje de pacientes con medicamentos específicos para la enfermedad pulmonar intersticial, entre los que tenían datos disponibles: 168 antes y 72 después, no los 209 del estudio completo.
Procedimientos invasivos planeados antes y realizados después de usar la herramienta
Bajó de 38.3% a 15.6% entre los pacientes con datos disponibles
El estudio fue observacional: los médicos la usaron cuando quisieron, y otras cosas pudieron influir.
Si a usted o a alguien de su familia le están investigando una enfermedad pulmonar llamada enfermedad pulmonar intersticial, sabrá que el camino es largo. Es un grupo de enfermedades que dejan cicatrices en los pulmones, y los médicos a veces tardan en saber cuál de todas es.
El problema es que varias de estas enfermedades se ven parecidas en las imágenes del pulmón, pero el tratamiento es distinto. La fibrosis pulmonar idiopática se trata con medicamentos antifibróticos; otras, con medicamentos que bajan las defensas. Confundirlas retrasa el tratamiento y a veces lleva a una biopsia de pulmón, que tiene riesgos.
Un estudio siguió a 209 pacientes en 47 centros médicos de Estados Unidos, entre julio de 2024 y junio de 2026. La mayoría eran mayores, con una edad media de 73 años. Los médicos usaron una herramienta de inteligencia artificial llamada Fibresolve, aprobada por la Administración de Alimentos y Medicamentos de Estados Unidos, que analiza las tomografías del tórax para apoyar el diagnóstico.
Entre los pacientes con datos disponibles, el uso de medicamentos para la enfermedad subió del 12.5% al 44.4% después de la prueba. Y los procedimientos invasivos, como la biopsia, bajaron: el 38.3% estaba planeado antes, y sólo el 15.6% se realizó después.
Hay que tener cuidado con esas cifras. No son de los 209 pacientes, sino de grupos más pequeños con información disponible: 168 antes y 72 después en el caso de los medicamentos; 154 y 77 en el de los procedimientos. Muchos pacientes no tenían datos de seguimiento, así que el tamaño real del beneficio es incierto.
Tampoco se puede decir que la herramienta causara los cambios. El estudio fue observacional: los médicos la usaron cuando quisieron, y otras cosas pudieron influir. No muestra que mejore la supervivencia ni la función pulmonar. Y como toda prueba, puede equivocarse, así que acompaña la evaluación del médico, no la reemplaza.
Si usted o un familiar enfrentan este diagnóstico, pregunte si una herramienta así puede formar parte de la evaluación. Y fíjese si en su clínica u hospital ofrecen revisión de imágenes con inteligencia artificial, y cómo podría cambiar su plan de tratamiento.
Qué significa para usted
El estudio no prueba que la herramienta causara esos cambios, así que conviene mirar el resultado como una señal y no como una certeza. Si a usted le están investigando una enfermedad pulmonar, pregunte en su clínica si usan inteligencia artificial para leer las tomografías y cómo ese resultado podría cambiar su tratamiento.
Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445
Quién pagó: Esta investigación no recibió financiación externa; IMVARIA, Inc. proporcionó datos de registro desidentificados y subsidiará el cargo por procesamiento del artículo, pero no financió a los autores ni participó en el diseño del estudio, el análisis de datos o la preparación del manuscrito.
No tome esto como consejo médico profesional.
Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1342 palabras · unos 7 minLeerla →Cerrar
Una herramienta de IA ya autorizada coincidió con más tratamiento y menos operaciones en pacientes con cicatrices pulmonares, según un estudio observacional en EE. UU.
El estudio no prueba causa, y muchos enfermos quedaron sin datos de seguimiento. Aun así, es el primer retrato de cómo se usa esta tecnología en la práctica real.
Cientos de personas llegan cada año a la consulta con cicatrices en el pulmón, y el camino hasta ponerle nombre a lo que tienen es largo. Los autores de este trabajo lo dicen sin adornos: el diagnóstico de la enfermedad pulmonar intersticial es difícil y suele llegar tarde, y hacen falta herramientas accesibles y poco invasivas que lo aceleren sin exponer al paciente a más riesgos1. La revisión recoge un estudio en el que, cuando el diagnóstico por fin se logra, ya pasaron meses o años, y ese tiempo perdido puede costar función pulmonar que no vuelve2.
La enfermedad pulmonar intersticial no es una sola cosa. Bajo ese nombre conviven varios subtipos que en la imagen del escáner pueden verse casi idénticos, pero que se tratan distinto. La revisión recoge un estudio en el que el más frecuente de los idiopáticos es la fibrosis pulmonar idiopática, que deja una supervivencia media de tres a cinco años tras el diagnóstico y, sin tratamiento, hace perder alrededor de 200 mililitros de capacidad pulmonar al año3. Por eso el nombre exacto importa tanto: no es una etiqueta, es lo que decide qué medicamento recibe alguien.
El problema es que la certeza diagnóstica no está repartida por igual. La revisión recoge un estudio en el que, pese a los avances enormes en tomografía, hasta el 61% de los pacientes con enfermedad pulmonar intersticial sigue siendo enviado a procedimientos invasivos —lavado broncoalveolar, biopsia por broncoscopio o biopsia quirúrgica de pulmón— para ganar confianza en el diagnóstico4. La revisión recoge un estudio en el que esos procedimientos no son gratuitos para el cuerpo: la biopsia quirúrgica de pulmón conlleva una mortalidad de al menos 2 a 3% en los 30 días posteriores a la operación5. Ahí está el nudo: se opera para saber, y a veces se opera de más.
La revisión recoge un estudio en el que Fibresolve es un programa de inteligencia artificial autorizado por la Administración de Alimentos y Medicamentos de Estados Unidos. Analiza las imágenes de tomografía del tórax y devuelve un texto binario: positivo, que sugiere fibrosis pulmonar idiopática, o negativo, que no respalda ese diagnóstico6. Está pensado como apoyo al diagnóstico, no como sustituto del médico7. El estudio que hoy contamos reunió a 209 pacientes en 47 centros de Estados Unidos —16 académicos y 31 comunitarios— entre julio de 2024 y junio de 2026, y observó qué pasó antes y después de usar la herramienta.
Lo que encontró: la proporción de pacientes que recibía medicamentos específicos para la enfermedad subió de 12.5% a 44.4% después de usar Fibresolve8. Y la proporción de pacientes con un procedimiento invasivo planeado o realizado bajó de 38.3% a 15.6%9. En los centros comunitarios, donde suele haber menos especialistas en pulmón, de 51 pacientes con datos disponibles solo tres, el 5.9%, terminó en un procedimiento invasivo tras la prueba10. Los autores también notaron algo que frena una lectura fácil: la capacidad pulmonar de los pacientes positivos y negativos era parecida, de modo que los positivos no eran simplemente los más enfermos11.
Conviene frenar aquí. Los porcentajes no salen de los 209 pacientes completos: el de medicamentos compara 168 pacientes antes con 72 después, y el de procedimientos compara 154 antes con 77 después. El propio trabajo reconoce que los datos faltantes —sobre tratamientos previos y posteriores, planificación de procedimientos y subtipo final— son su limitación principal12. Y hay algo más de fondo: como el estudio solo observó lo que ocurría, sin asignar a nadie al azar, no puede probar que la herramienta haya causado los cambios; los autores lo escriben con todas las letras13.
Así lo leemos nosotros. Cuando el especialista queda lejos o la lista de espera es larga, mucha gente deja de exigir y se conforma con lo que tiene cerca. Si eso describe su caso, la aparición de una herramienta que entrega un resultado claro y rápido puede empujar a los médicos generales a decidirse antes, sin esperar a que llegue el experto. Lo que esperaríamos ver, entonces, es que en los centros comunitarios —justamente donde no hay especialistas a mano— los médicos traten más y operen menos. Sabríamos que nos equivocamos si viéramos lo contrario: médicos comunitarios igual de paralizados, sin iniciar tratamiento ni evitar procedimientos hasta que aparezca el experto, aunque la herramienta ya les haya dado su respuesta. Si a usted o a un familiar le están estudiando un problema de pulmón y el neumólogo queda lejos, pregunte si existe alguna herramienta ya autorizada que ayude a su médico de cabecera a orientar el diagnóstico mientras llega esa consulta.
Una segunda cosa que nos hace ruido: cuando una máquina se presenta como limpia, sin cansancio ni preferencias, es tentador tratarla como veredicto y dejar de lado el propio juicio o la conversación entre colegas. Esperaríamos que algunos médicos usen el resultado como un dato más —contrastado con la historia clínica, con la imagen y con otros colegas— y que otros lo tomen como palabra final, sobre todo cuando coincide con lo que ya sospechaban. Nos equivocaríamos si en la práctica viéramos siempre lo primero. Usted puede preguntar quién más revisó su caso además de la herramienta, y si el diagnóstico se discutió entre varios médicos. Un resultado de software no reemplaza esa conversación.
Vale la pena mirar qué otras herramientas parecidas existen. La revisión recoge un estudio en el que Fibresolve recibió en enero de 2024 la primera autorización de la FDA para un software de inteligencia artificial con uso diagnóstico en esta enfermedad, y desde entonces su uso clínico creció rápido14. La revisión recoge un estudio en el que se mencionan dos plataformas más autorizadas después, con indicaciones distintas: una cuantifica la pérdida de volumen pulmonar en pacientes ya diagnosticados, y otra detecta un patrón radiológico concreto y avisa a especialistas de centros de enfermedad intersticial, sin dar un diagnóstico clínico. La revisión recoge un estudio en el que ninguna de esas dos entrega directamente un diagnóstico, y sobre ellas no hay todavía datos de impacto en la práctica real.
Otros trabajos que leímos apuntan en la misma dirección, aunque en otros terrenos. Son familias de hallazgos distintas, y ninguna de ellas dice nada sobre los pulmones de nadie en particular.
Lo que falta es lo más importante. Los propios autores escriben que harían falta datos a lo largo del tiempo sobre pruebas de función pulmonar, hospitalizaciones y supervivencia, comparando pacientes con y sin esta prueba, para confirmar si su uso mejora de verdad los resultados a largo plazo15. Hoy no existe esa respuesta. Tampoco hay datos fuera de Estados Unidos, y el estudio no evaluó si la herramienta acierta en subtipos que no sean la fibrosis idiopática.
Sobre quién paga y quién opina, conviene ser claros. La empresa fabricante, IMVARIA, entregó los datos ya anonimizados para este trabajo y costeará la tarifa de publicación del artículo, aunque no financió a los autores ni participó en el diseño, el análisis ni la redacción16. Uno de los autores declara haber integrado consejos asesores de varias farmacéuticas y ser conferencista de otra; otra autora declara una relación con una academia médica que incluye honorarios por conferencias16.
Aquí está lo que usted puede hacer con todo esto. Si le hablan de cicatrices en el pulmón, pregunte en qué etapa está y qué se puede hacer ahora, sin esperar a que empeore, y pida que le expliquen cómo el resultado de cualquier estudio se aplica a su caso concreto y no al de un paciente promedio. Pregunte también si su diagnóstico se discutió entre varios médicos y no solo con una herramienta. Nada de esto significa saltarse pasos ni retrasar una consulta: significa llegar a ella con mejores preguntas.
¿Sabe usted qué especialista le corresponde y a qué distancia queda?
De dónde sale cada dato de contexto, y cuánto leímos de cada documento
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The diagnosis of interstitial lung disease (ILD) is challenging and frequently delayed. Clinically accessible and minimally invasive diagnostic tools are needed to expedite the diagnosis of ILD while minimizing risk to patients."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Unfortunately, even when the correct diagnosis is made, it is frequently delayed by months to years, and time lost to diagnostic delays can lead to irreversible loss of lung function"
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "IPF carries a median survival of only 3–5 years after diagnosis, and left untreated leads to an average loss of approximately 200 mL in forced vital capacity (FVC) annually."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Despite the substantial advances in imaging techniques, up to 61% of patients with ILD are still referred for invasive procedures—such as bronchoalveolar lavage, transbronchial biopsy, or surgical lung biopsy—to improve diagnostic confidence"
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Surgical lung biopsy is considered even higher-risk than using a bronchoscopic approach, with 30-day post-operative mortality of at least 2–3%."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Fibresolve produces a binary text output: positive (suggestive of IPF) or negative (unsupportive of IPF)."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Fibresolve (Version 1, IMVARIA, Inc., Berkeley, CA, USA) is a novel, recently FDA-authorized AI software for the diagnosis of ILD. Specifically, Fibresolve was designed to provide a diagnostic subtype classification in suspected cases of ILD and serve as an adjunct in the diagnosis of IPF."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The percentage of patients receiving ILD-specific pharmacotherapy increased significantly following Fibresolve utilization (12.5% to 44.4%, p < 0.001)."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Pre-Fibresolve, 59/154 (38.3%) patients had an invasive diagnostic procedure planned; post-Fibresolve, only 12/77 (15.6%) patients had an invasive procedure performed ( p < 0.001)."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Of 51 patients with available data, only three patients (5.9%) at community-based centers underwent an invasive diagnostic procedure following Fibresolve testing."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The median percent predicted FVC was not different in patients with a positive vs. negative Fibresolve test result, illustrating that positive results were not simply limited to patients with more advanced fibrotic disease."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This early clinical utility study was limited by missing data, specifically related to pre- and post-test medical therapies, procedural planning, and final ILD subtype designation."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Though this study describes trends in clinical practice patterns before and after Fibresolve utilization, the observational study design limits inference of causality."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "In January 2024, Fibresolve received the first FDA authorization of an AI software for diagnostic use in ILD, and since that time, clinical utilization has rapidly increased"
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Ultimately, longitudinal data on pulmonary function tests, hospitalization rates, and survival compared between patients with and without Fibresolve testing would confirm whether Fibresolve use leads to improvement in long-term clinical outcomes for patients."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - el artículo del que trata esta nota — el artículo completo — el pasaje: "IMVARIA, Inc. provided de-identified registry data for this manuscript. IMVARIA, Inc. will subsidize the manuscript’s Article Processing Charge, but it did not fund the authors or participate in study design, data analysis, or manuscript preparation."
Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445
Quién pagó: Esta investigación no recibió financiación externa; IMVARIA, Inc. proporcionó datos de registro desidentificados y subsidiará el cargo por procesamiento del artículo, pero no financió a los autores ni participó en el diseño del estudio, el análisis de datos o la preparación del manuscrito.
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 · Diagnostics (Basel, Switzerland) · the paper, 3 Aug 2026 · free
An AI Tool Read the Lung Scans. More Patients Started Treatment, and Fewer Had Surgery.
In a study of 209 patients in the United States, more people began lung treatment and fewer biopsies were done after the tool was used. The study cannot prove the tool caused it.
Short version · the longer version follows, about 8 min
Ask the weeklyAI chat
Ask me about this study: who was studied, what it found, and what it does not say.
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
- patients being checked for lung scarring (interstitial lung disease)
- How many
- 209 patients
- Where
- United States
- When
- July 2024 to June 2026
- Kind of study
- analysis of what people did
- Who did it
- Northwestern University, Indiana University, Thomas Jefferson University, Massachusetts General Hospital, and an independent researcher
- The limit that matters
- Only some patients had follow-up data, and the study cannot prove the tool caused the changes.
These are shares of patients in a US study of 209 people; only some patients had follow-up information (72 for the medicine figures, 77 for the procedures), and the study cannot prove the tool caused the changes.
What happened before and after the AI tool was used
More patients were on lung medicine after the tool was used (12.5% before, 44.4% after)
Fewer patients had an invasive procedure after the tool was used (38.3% planned before, 15.6% performed after)
You or someone close to you is being checked for interstitial lung disease, a group of conditions that scar the lungs. Doctors call it ILD. It is hard to pin down, and the wait for an answer can stretch for months.
Different ILD types can look alike on a chest scan, yet they need different treatments. One type, idiopathic pulmonary fibrosis, is treated with antifibrotic medicine. Others are treated with drugs that calm the immune system. So telling them apart matters, and doctors often can't agree on what a scan shows.
Researchers from several US medical schools and an independent researcher published a study in the journal Diagnostics. They followed 209 patients at 47 medical centers across the United States from July 2024 to June 2026. Sixteen centers were academic, and 31 were based in the community. The median age was 73.
During their workup, doctors used Fibresolve, an imaging software tool authorized by the US Food and Drug Administration. It reads chest CT scans and returns one of two words: positive, meaning the scan suggests IPF, or negative, meaning it does not support IPF. Of the 209 patients, 89 tested positive, a little over four in ten.
After the tool was used, more patients were on lung medicine: about 12.5 percent before, about 44 percent after. And fewer went through invasive procedures like lung biopsy: about 38 percent had one planned beforehand, and about 16 percent had one performed afterward.
One caution sits inside those numbers. Only some patients had follow-up information, 72 for the medicine figures and 77 for the procedures. So the exact size of the change is uncertain. And this was an observation of what doctors did, not a test that compared two groups. The study cannot prove the tool caused any of it. It also says nothing yet about whether patients live longer or keep more lung function. The people studied were all in the United States.
The tool can also be wrong. The article describes earlier studies in which the tool was good at ruling IPF in when it was there, but it missed many cases. A false positive could lead to the wrong medicine.
If a lung workup comes your way, ask whether an imaging tool like this one is available where you are being seen, and whether it could be part of the evaluation.
What this means for you
What the study cannot show is whether patients lived longer or kept more lung function, and it says nothing about care outside the United States. So ask the doctor handling your lung workup whether a tool like this one is available where you are being seen, and whether it could be part of your evaluation. Watch for whether a biopsy is proposed, and whether the reason for it is explained to you.
Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445
Who paid: This research received no external funding; IMVARIA, Inc. provided de-identified registry data and will subsidize the article processing charge but did not fund the authors or participate in study design, data analysis, or manuscript preparation.
Do not take this as professional medical advice.
The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.
The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1512 words · about 8 minRead it →Close
An AI Read the Scans. Then More Patients Started Treatment.
In a 209-patient US study, a lung-scarring tool was followed by more prescriptions and fewer biopsies — but the study can’t prove the tool did it.
Interstitial lung disease is a family of conditions that scar the lungs, and getting the right name on the right subtype is slow and often wrong-footed. Diagnosing it is hard and frequently delayed, and the authors write plainly that accessible, minimally invasive tools are needed to speed it up while lowering risk to patients1. The article describes a study in which, even when the correct diagnosis is reached, it is frequently delayed by months to years, and that lost time can lead to irreversible loss of lung function2.
The stakes are sharpest in idiopathic pulmonary fibrosis, or IPF, the most common idiopathic subtype. The article describes a study in which IPF carries a median survival of only 3–5 years after diagnosis, and left untreated it leads to an average loss of roughly 200 mL in forced vital capacity each year3. That is a measurement of how much air a person can force out of their lungs in one breath — about a small glass of liquid, given up annually. The article notes that current treatments can slow progression, but no therapy has been shown to improve lung function in IPF patients.
The article describes a study in which the gold standard for sorting out which ILD a person has is a multidisciplinary discussion at an expert center, where clinical history, laboratory data, chest imaging and, when available, tissue samples are reviewed together1. But that kind of expert panel sits mostly at academic centers, and waits can be long. So many patients are first seen in community clinics by primary care doctors and general lung specialists, who the article says may be less comfortable recognizing and managing these rare diseases.
One consequence shows up in the numbers: the article describes a study in which, despite big advances in CT imaging, up to 61% of patients with ILD are still referred for invasive procedures — bronchoalveolar lavage, transbronchial biopsy, or surgical lung biopsy — to raise diagnostic confidence4. Those procedures carry real risk. The article describes a study in which surgical lung biopsy has a 30-day post-operative mortality of at least 2–3%5. And the tissue they retrieve often doesn’t settle the question: the article describes a study in which the diagnostic yield for cryobiopsy and traditional forceps biopsy in diffuse lung disease was recently reported at 72% and 62.5%.
Here is the mechanism, in plain words. Fibresolve (Version 1, IMVARIA, Inc., Berkeley, CA, USA) is described in the article as a novel, recently FDA-authorized AI software for ILD diagnosis, designed to give a diagnostic subtype classification in suspected ILD cases and to serve as an adjunct in diagnosing IPF6. The article describes a study in which it analyzes chest CT images using a deep learning algorithm — a convolutional neural network — and, unlike other IPF classifier tools, its analysis is not limited to the pre-defined features of usual interstitial pneumonia6. It produces a binary text output: positive, meaning suggestive of IPF, or negative, meaning unsupportive of IPF7.
The article describes a study in which, in January 2024, Fibresolve received the first FDA authorization of an AI software for diagnostic use in ILD, and clinical utilization has risen rapidly since8. Two more AI platforms have since been authorized for ILD — e-Lung and IQ-UIP — but the article notes that neither directly provides a clinical diagnosis, and data on their impact in real-world practice are lacking.
The new study is a prospective, observational look at how the tool behaved in ordinary practice across the United States between July 2024 and June 2026. Two hundred nine patients underwent evaluation with Fibresolve across 47 centers: 16 academic centers with 57 patients, and 31 community-based centers with 152 patients. The median age was 73, 83 patients (39.7%) were female, and the median percent predicted FVC was 84%. Fibresolve came back positive in 89 of 209 patients, or 42.6%.
Two changes followed. The percentage of patients receiving ILD-specific pharmacotherapy rose from 12.5% to 44.4%9. And invasive diagnostic procedures fell: before Fibresolve, 59 of 154 patients (38.3%) had one planned; afterward, only 12 of 77 (15.6%) had one performed10. Both differences were statistically significant.
The article is careful about what those figures cover. The 12.5%-to-44.4% comparison rests on 168 patients with pre-test data and 72 with post-test data; the procedure comparison rests on 154 and 77. The study was limited by missing data on pre- and post-test therapies, procedural planning, and final ILD subtype11. And because it was observational, the design limits any inference of causality12. The authors also flag selection bias — Fibresolve was used at physician discretion, not assigned — plus possible residual confounding and observational bias, and they note a false-positive result could lead to starting antifibrotic therapy wrongly if the correct diagnosis is never established another way.
One finding pushes against an easy assumption. The median percent predicted FVC was not different between patients with a positive versus a negative result, which the article says shows positive results were not simply limited to people with more advanced fibrotic disease13. In other words, the tool was not just confirming what advanced disease already told doctors.
Where the tool was used seemed to matter. At community-based centers, of 51 patients with available data, only three (5.9%) underwent an invasive diagnostic procedure after Fibresolve testing14. Academic centers kept doing more procedures. The authors offer reasons without settling the question: possibly harder cases at academic centers, a higher pre-existing procedure rate there, or greater confidence in the result at community clinics.
Who ran this matters to how you read it. The authors are Arjun Prakash Tambe and Kavitha C. Selvan of Northwestern University, with colleagues from Indiana University, Thomas Jefferson University, Massachusetts General Hospital and an independent researcher. IMVARIA, Inc. provided de-identified registry data for the manuscript and will subsidize the article processing charge, but the article states it did not fund the authors or take part in study design, data analysis or manuscript preparation15. One author reports advisory-board roles with several drug companies and a speaking role with Boehringer Ingelheim; another reports a relationship with the American Academy of Chest Physicians including speaker fees; the remaining four declare no conflicts.
Here is how we read it. When an expert skill is scarce, the people farthest from the expert centers tend to get the least confident version of the service, and the gap often shows up not as an obvious failure but as more guesswork and more aggressive testing. That is the pattern this study sits inside: a tool that hands a small, readable answer to a clinic that has no expert panel down the hall. What we expect, if that reading is right, is that the change in behavior should be largest where the specialist is missing — and that is roughly what the figures show, with community clinics dropping procedures far more sharply than teaching hospitals. We could be wrong. If community clinics were ordering just as many invasive procedures after the tool as before, this reading would collapse.
We also read a quieter question into it. A tool that returns “positive” or “negative” can end up carrying a decision that used to belong to a conversation between a doctor and a patient, and the person whose lungs are being described may never learn that a machine result was part of the plan. We would want to know whether clinics say so out loud, and whether there is an easy way to ask for a second look. You can ask, in plain words, whether a computer result was part of the decision and who reviewed it — and ask that the reasoning be written into your record.
What this study does not show is whether any of it changed how people actually fare. The article is explicit that longitudinal data on pulmonary function tests, hospitalization rates and survival, compared between patients with and without Fibresolve testing, would be needed to confirm whether the tool improves long-term outcomes16. Until then, the honest summary is that the tool was used alongside more treatment and fewer procedures — not that it caused either.
So what does this leave you with, wherever you are reading this? If you or someone close to you is being worked up for lung scarring, the useful move is not to ask for a machine and not to refuse one. It is to ask what has actually been run on the scans, what the result said, and what the plan is if that result is positive and what the plan is if it is negative. Ask whether your case has been reviewed by a team that sees this disease often, and if not, how long the wait is. Ask whether a biopsy is being proposed to answer a question that imaging might already answer. Those questions cost nothing, and every one of them is a question a doctor can answer.
Where each piece of context comes from, and how much of it we read
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "The diagnosis of interstitial lung disease (ILD) is challenging and frequently delayed. Clinically accessible and minimally invasive diagnostic tools are needed to expedite the diagnosis of ILD while minimizing risk to patients."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Unfortunately, even when the correct diagnosis is made, it is frequently delayed by months to years, and time lost to diagnostic delays can lead to irreversible loss of lung function"
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "IPF carries a median survival of only 3–5 years after diagnosis, and left untreated leads to an average loss of approximately 200 mL in forced vital capacity (FVC) annually."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Despite the substantial advances in imaging techniques, up to 61% of patients with ILD are still referred for invasive procedures—such as bronchoalveolar lavage, transbronchial biopsy, or surgical lung biopsy—to improve diagnostic confidence"
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Surgical lung biopsy is considered even higher-risk than using a bronchoscopic approach, with 30-day post-operative mortality of at least 2–3%."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Fibresolve (Version 1, IMVARIA, Inc., Berkeley, CA, USA) is a novel, recently FDA-authorized AI software for the diagnosis of ILD. Specifically, Fibresolve was designed to provide a diagnostic subtype classification in suspected cases of ILD and serve as an adjunct in the diagnosis of IPF."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Fibresolve produces a binary text output: positive (suggestive of IPF) or negative (unsupportive of IPF)."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "In January 2024, Fibresolve received the first FDA authorization of an AI software for diagnostic use in ILD, and since that time, clinical utilization has rapidly increased"
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "The percentage of patients receiving ILD-specific pharmacotherapy increased significantly following Fibresolve utilization (12.5% to 44.4%, p < 0.001)."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Pre-Fibresolve, 59/154 (38.3%) patients had an invasive diagnostic procedure planned; post-Fibresolve, only 12/77 (15.6%) patients had an invasive procedure performed ( p < 0.001)."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "This early clinical utility study was limited by missing data, specifically related to pre- and post-test medical therapies, procedural planning, and final ILD subtype designation."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Though this study describes trends in clinical practice patterns before and after Fibresolve utilization, the observational study design limits inference of causality."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "The median percent predicted FVC was not different in patients with a positive vs. negative Fibresolve test result, illustrating that positive results were not simply limited to patients with more advanced fibrotic disease."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Of 51 patients with available data, only three patients (5.9%) at community-based centers underwent an invasive diagnostic procedure following Fibresolve testing."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "IMVARIA, Inc. provided de-identified registry data for this manuscript. IMVARIA, Inc. will subsidize the manuscript’s Article Processing Charge, but it did not fund the authors or participate in study design, data analysis, or manuscript preparation."
- Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445 - the article this story is about — the whole article — the passage: "Ultimately, longitudinal data on pulmonary function tests, hospitalization rates, and survival compared between patients with and without Fibresolve testing would confirm whether Fibresolve use leads to improvement in long-term clinical outcomes for patients."
Tambe, A. P., Boente, R. D., George, G. et al. (2026). Clinical Utility of an FDA-Authorized Artificial Intelligence Imaging Platform in Interstitial Lung Disease Diagnosis. Diagnostics. https://doi.org/10.3390/diagnostics16152445
Who paid: This research received no external funding; IMVARIA, Inc. provided de-identified registry data and will subsidize the article processing charge but did not fund the authors or participate in study design, data analysis, or manuscript preparation.
Do not take this as professional medical advice.
The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.