Connect with us

Health

Researchers Develop AI Method That Makes Computer Vision Models More Explainable

A new technique developed by MIT researchers could help make artificial intelligence systems more accurate and transparent in high-stakes fields such as health care and autonomous driving by improving how computer vision models explain their decisions.

Published

on

MIT researchers have developed a new explainable AI method that improves the accuracy and transparency of computer vision models, helping users trust AI predictions in healthcare and autonomous driving.
Image credit: Tara/Pexels

MIT researchers have developed a new explainable AI method that improves the accuracy and transparency of computer vision models, helping users trust AI predictions in healthcare and autonomous driving.

Researchers at MIT have developed a new approach to make computer vision models more transparent, offering a potential boost to trust and accountability in safety-critical applications such as medical diagnosis and autonomous driving.

In a media statement, the researchers said the method improves on a widely used explainability technique known as concept bottleneck modeling, which enables AI systems to show the human-understandable concepts behind a prediction. The new approach is designed to produce clearer explanations while also improving prediction accuracy.

Why explainable AI matters

In areas such as health care, users often need more than just a model’s output. They want to understand why a system arrived at a particular conclusion before deciding whether to rely on it. Concept bottleneck models attempt to address that need by forcing an AI system to make predictions through a set of intermediate concepts that humans can interpret.

For example, when analysing a medical image for melanoma, a clinician might define concepts such as “clustered brown dots” or “variegated pigmentation.” The model would first identify those concepts and then use them to arrive at its final prediction.

But the researchers said pre-defined concepts can sometimes be too broad, irrelevant or incomplete for a specific task, limiting both the quality of explanations and the model’s performance. To overcome that, the MIT team developed a method that extracts concepts the model has already learned during training and then compels it to use those concepts when making decisions.

The approach relies on two specialised machine-learning models. One extracts the most relevant internal features learned by the target model, while the other translates them into plain-language concepts that humans can understand. This makes it possible to convert a pretrained computer vision model into one capable of explaining its reasoning through interpretable concepts.

“In a sense, we want to be able to read the minds of these computer vision models. A concept bottleneck model is one way for users to tell what the model is thinking and why it made a certain prediction. Because our method uses better concepts, it can lead to higher accuracy and ultimately improve the accountability of black-box AI models,” Antonio De Santis, lead author of the study, said in a media statement.

De Santis is a graduate student at Polytechnic University of Milan and carried out the research while serving as a visiting graduate student at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). The paper was co-authored by Schrasing Tong, Marco Brambilla of Polytechnic University of Milan, and Lalana Kagal of CSAIL. The research will be presented at the International Conference on Learning Representations.

Concept bottleneck models have gained attention as a way to improve AI explainability by introducing an intermediate reasoning step between an input image and the final output. In one example, a bird-classification model might identify concepts such as “yellow legs” and “blue wings” before predicting a barn swallow.

However, the researchers noted that these concepts are often generated in advance by humans or large language models, which may not always match the needs of the task. Even when a model is given a fixed concept set, it can still rely on hidden information not visible to users, a challenge known as information leakage.

“These models are trained to maximize performance, so the model might secretly use concepts we are unaware of,” De Santis said in a media statement.

The team’s solution was to tap into the knowledge the model had already acquired from large volumes of training data. Using a sparse autoencoder, the method isolates the most relevant learned features and reconstructs them into a small number of concepts. A multimodal large language model then describes each concept in simple language and labels the training images by marking which concepts are present or absent.

The annotated dataset is then used to train a concept bottleneck module, which is inserted into the target model. This forces the model to make predictions using only the extracted concepts.

The researchers said one of the biggest challenges was ensuring that the automatically identified concepts were both accurate and understandable to humans. To reduce the risk of hidden reasoning, the model is limited to just five concepts for each prediction, encouraging it to focus only on the most relevant information and making the explanation easier to follow.

When tested against state-of-the-art concept bottleneck models on tasks including bird species classification and skin lesion identification, the new method delivered the highest accuracy while also producing more precise explanations, according to the researchers. It also generated concepts that were more relevant to the images in the dataset.

Still, the team acknowledged that the broader challenge of balancing accuracy and interpretability remains unresolved.

“We’ve shown that extracting concepts from the original model can outperform other CBMs, but there is still a tradeoff between interpretability and accuracy that needs to be addressed. Black-box models that are not interpretable still outperform ours,” De Santis said in a media statement.

Looking ahead, the researchers plan to explore ways to further reduce information leakage, possibly by adding additional concept bottleneck modules. They also aim to scale up the method by using a larger multimodal language model to annotate a larger training dataset, which could improve performance further.

This latest work adds to growing efforts to make AI systems not only more powerful, but also more understandable in domains where trust can be as important as accuracy.

EP Staff is the editorial team at EdPublica, an independent media organisation focused on science, education, environment and public policy. The team produces evidence-based news, features, explainers and analysis on issues that shape society and everyday life.

Society

Intellectual disabilities linked to severe educational disadvantage in India, study finds

A new study published in Scientific Reports highlights significant educational disadvantages among people with intellectual and developmental disabilities and mental illness in India. Alongside school enrolment and infrastructure data, the findings underline persistent gaps in inclusive education, gender equality and access to learning support.

Published

on

Two men smiling and interacting during a conversation, highlighting inclusion and social interaction.
Meaningful social interaction and inclusion are essential to building a society that respects and supports people with disabilities.Image credit: Unsplash

New Delhi: People with intellectual and developmental disabilities (IDD) in India face substantially greater educational disadvantages than those without disabilities, with gender and economic inequalities further widening the gap, according to a study published in Scientific Reports on October 15, 2026.

The research, analysed data from 402,994 adults using the 76th round of the National Sample Survey (NSS), conducted in 2018.

Researchers found that individuals with intellectual and developmental disabilities had a 71.3% predicted probability of receiving no formal education. The corresponding probability among people reporting mental illness was 48.3%.

The study also found that individuals with IDD had 28.9 times higher odds of receiving no formal education than those without disabilities. For individuals with mental illness, the odds were 3.3 times higher.

The findings highlight how disability-related barriers can translate into long-term educational exclusion, particularly when combined with poverty and gender discrimination.

Gender and economic inequalities compound exclusion

The researchers examined how educational disadvantages varied across gender and household economic groups.

Women with mental illness faced a particularly pronounced disadvantage. They were 18.7 percentage points more likely than men with mental illness to have received no formal education.

Economic circumstances also influenced educational outcomes. Among individuals with IDD, the predicted probability of receiving no formal education declined from approximately 79.9% in the poorest household group to 62.9% in the richest group.

For individuals with mental illness, the corresponding probability declined from approximately 59.9% to 33.5%.

However, economic advantage did not eliminate gender disparities. The researchers found that women with mental illness continued to experience educational disadvantages even in the wealthiest household category.

The study distinguishes between intellectual and developmental disabilities and mental illness, highlighting the importance of addressing their different educational and support requirements rather than treating them as a single category.

Enrolment figures reveal the scale of the challenge

The research findings gain significance when considered alongside India’s school-enrolment figures.

According to the Unified District Information System for Education Plus (UDISE+) data cited in a Lok Sabha response on March 30, 2026, approximately 1.91 million children with special needs were enrolled in government, government-aided and local-body schools during 2024–25.

The corresponding figure was approximately 1.93 million in 2023–24. Girls accounted for 827,521 of the enrolled children in 2024–25, representing approximately 43.4% of the total.

The figures indicate the scale of the population requiring inclusive educational support. However, enrolment alone does not establish whether children are receiving appropriate learning assistance, participating fully in classrooms or progressing through different levels of education.

Intellectual
disability
India

Representational image.Image credit:Unsplash

Infrastructure and specialist support remain important components of inclusive schooling. The Ministry of Education’s Annual Report 2024–25 recorded that 11.34 lakh schools had ramps, 7.69 lakh had ramps with handrails and 5.06 lakh had children-with-special-needs-friendly toilets during 2023–24.

The government also reported 52,274 government, government-aided and local-body schools with special educators in 2024–25. These figures point to the need to assess not only school access but also the availability and distribution of trained personnel and accessible facilities.

Strengthening inclusive education

India’s Right of Children to Free and Compulsory Education Act, 2009, and the Rights of Persons with Disabilities Act, 2016, provide the legal framework for improving educational access for children with disabilities.

Under the Samagra Shiksha scheme, the government provides support through early identification, assessment, assistive devices, therapeutic interventions, transport allowances, resource support and individualised learning assistance.

The scheme also provides a stipend of Rs 200 per month for ten months to eligible girls with special needs to encourage school participation.

However, the research highlights the importance of addressing barriers that extend beyond school infrastructure and financial assistance. Teacher preparedness, social stigma, gender discrimination and inadequate access to specialised support can influence educational participation and attainment.

The researchers emphasise the need for gender-sensitive and income-sensitive interventions, including school-based mental-health support, community awareness programmes and stronger access to healthcare services.

The study also acknowledges limitations arising from its reliance on cross-sectional and self-reported data. Its findings establish associations rather than proving that disability alone causes educational exclusion. Moreover, because the analysis covers adults, it cannot directly measure the present-day enrolment or dropout rates of children with disabilities.

Nevertheless, the findings offer an important indication of how educational inequalities can persist into adulthood. They underline the need to strengthen inclusive education systems so that children with disabilities are not merely enrolled in schools but are also supported in completing their education and accessing further opportunities.

Continue Reading

Health

TB deaths decline in India, but diagnostic gaps threaten elimination goals

India has reduced tuberculosis mortality by 38% since 2015, according to the WHO Global Tuberculosis Report 2026. However, gaps in diagnostic infrastructure, particularly in drug-resistance testing and laboratory facilities, continue to challenge the country’s TB elimination efforts.

Published

on

Healthcare professional examining a chest X-ray to assess lung abnormalities.
A healthcare professional examines a chest radiograph as part of the diagnostic process for lung diseases, including tuberculosis.Image credit:Pixels

India has recorded a 38% decline in tuberculosis (TB) mortality since 2015, outperforming the global reduction of 35%. However, significant gaps in diagnostic infrastructure, particularly at primary healthcare facilities, could complicate the country’s efforts to eliminate the disease.

The findings present a mixed picture of India’s TB response. The World Health Organization’s (WHO) Global Tuberculosis Report 2026, released on November 12, documents improvements in mortality and treatment outcomes. At the same time, a multi-state study published in the journal Microbiology Spectrum in August 2026 identifies substantial shortages in essential diagnostic services, including facilities required to detect drug-resistant TB.

According to the WHO report, India’s estimated TB mortality rate declined from 28 deaths per lakh population in 2015 to 18 per lakh in 2025. The country also recorded a 94% treatment success rate among patients with drug-susceptible TB.

The figures indicate progress in early detection, treatment access and patient support. However, reducing mortality does not necessarily mean that diagnostic services are equally accessible across the country’s healthcare system.

Diagnostic infrastructure remains inadequate

The study, titled Evaluating the nationwide availability of essential diagnostics for tuberculosis toward achieving a TB-free India, examined 327 public healthcare facilities across seven states and Delhi. It assessed the availability of diagnostic services recommended under India’s National Essential Diagnostics List.

One of its major findings was a 61.56% shortfall in culture and drug-susceptibility testing (CDST) laboratory capacity compared with population-based requirements. The researchers identified only 97 CDST laboratories against 6,308 TB Treatment Units.

Drug-susceptibility testing is essential for identifying whether tuberculosis bacteria are resistant to particular medicines. Without timely testing, patients with drug-resistant TB may face delays in receiving appropriate treatment, potentially increasing the risk of continued transmission and unsuccessful treatment outcomes.

The study also found that only 4.94% of the surveyed Primary Health Centres (PHCs) had the complete set of essential TB diagnostic facilities recommended for that level of healthcare.

Across the assessed facilities, cartridge-based nucleic acid amplification testing (CBNAAT), which can rapidly detect TB and certain forms of drug resistance, was available in 51.28%. X-ray services were also available in 51.28% of the facilities.

The researchers emphasised that expanding access to individual diagnostic technologies would not be sufficient without trained personnel, functional laboratory networks, maintenance, quality assurance and effective referral systems.

The findings should be interpreted as a multi-state assessment rather than a nationally representative survey of every public healthcare facility. Nevertheless, they identify infrastructure weaknesses that require further attention.

Screening expands across the country

India has expanded its screening initiatives under the TB Mukt Bharat Abhiyaan, which focuses on identifying undiagnosed cases, particularly among vulnerable populations.

According to the Union Ministry of Health and Family Welfare, more than 25 crore people were brought under intensified screening activities between December 2024 and November 8, 2026. The programme conducted over 11 crore chest X-rays and more than three crore molecular diagnostic tests.

TB
India
Diagnosis

A healthcare professional examining a chest X-ray.Image credit:Pexel

The campaign identified 48.64 lakh new TB cases, including 14.97 lakh asymptomatic cases. Detecting individuals without noticeable symptoms is particularly important because undiagnosed infections can delay treatment and contribute to transmission.

The government has also expanded nutritional assistance through the Ni-kshay Poshan Yojana and community support through the Ni-kshay Mitra initiative. These interventions recognise that TB treatment involves more than medication, particularly for patients experiencing undernutrition and financial difficulties.

Drug resistance and the challenge ahead

Drug-resistant tuberculosis remains an important concern for public health systems. Treating drug-resistant infections requires accurate diagnosis and appropriate treatment regimens, making access to comprehensive laboratory services essential.

The diagnostic study argues that rapid molecular tests must be complemented by functional culture-based testing facilities capable of identifying a broader range of drug resistance. It also highlights the potential of next-generation sequencing for resistance profiling, although costs, infrastructure and bioinformatics capacity remain barriers to wider implementation.

The WHO report, which draws on information from 187 countries and territories covering more than 99% of the global population and TB cases, provides an important assessment of progress towards international TB targets.

India’s declining mortality and improved treatment outcomes represent significant gains. However, sustaining this progress will depend on whether diagnostic services can reach patients before the disease advances and whether healthcare facilities can identify drug resistance promptly.

As India continues its TB elimination efforts, strengthening primary healthcare laboratories, expanding molecular testing and improving access to drug-susceptibility testing will be critical. The next phase of progress must focus not only on screening more people but also on ensuring that every suspected case can access accurate diagnosis and appropriate treatment.

Continue Reading

Space & Physics

Can Studying Cancer Tumours in Space Help Improve Chemotherapy on Earth?

Researchers are studying miniature models of colorectal cancer aboard the International Space Station to understand whether microgravity changes how cancer cells respond to chemotherapy.

Published

on

International Space Station with its large solar arrays extended, photographed against the darkness of space.
The International Space Station serves as a research laboratory for studying cancer cells, human health and biological processes in microgravity. Source: ISS/ NASA

Two people diagnosed with the same type of cancer may not respond to the same treatment in the same way. A chemotherapy drug that works well for one patient may be less effective for another. This happens partly because tumours differ in their genetic makeup, growth patterns and response to medicines. Understanding these differences is vital for developing cancer research and treatments that work better for individual patients.

By sending miniature models of cancer tumours into space, scientists at Wake Forest Institute for Regenerative Medicines in the United States are exploring an unusual way to explore the problem. They are investigating whether the absence of normal gravity changes how colorectal cancer cells respond to chemotherapy. The project, called Tumor Avatars in Space (TAVIS), uses the International Space Station (ISS) as a laboratory to study how cancer cells behave in microgravity.

The experiment uses organoids, which are small, three-dimensional models of human tissues grown in laboratories. For this project, researchers developed tumour organoids using cancer cells collected from patients with colorectal cancer. These miniature models retain some of the biological features of the original tumours, allowing scientists to study how they respond to medicines.

The researchers will compare how these tumour models respond to chemotherapy in space and on Earth. By studying the differences, they hope to understand whether microgravity changes the way cancer cells react to treatment.

Why Study Cancer Cells in Space?

Gravity influences how cells grow and arrange themselves. On Earth, cells are affected by gravitational forces and the way they settle within their surroundings. In microgravity, these conditions change, allowing cells to form three-dimensional structures in different ways. This makes space an interesting environment for studying how cells behave. Researchers want to know whether changes in these conditions can affect tumour growth, cell activity and responses to cancer drugs.

Cancer tumours are made up of more than just cancer cells. They contain different types of cells that interact with one another and their surroundings. These interactions can influence how a tumour grows and responds to treatment. Studying three-dimensional tumour models can help scientists understand some of these processes better.

In the TAVIS experiment, researchers will examine whether colorectal cancer cells become more sensitive to chemotherapy under microgravity conditions. If they find differences, they will investigate the genetic and cellular changes responsible for them. One area of interest is how cancer cells use energy. This process, known as metabolism, helps cells grow, survive and carry out their functions. Changes in metabolism could influence how effectively chemotherapy destroys cancer cells.

The researchers hope to identify the biological processes behind any changes in treatment response. They will then investigate whether similar effects can be reproduced on Earth, potentially by targeting particular genes or cellular processes.

What are Tumour Organoids?

Organoids are miniature, three-dimensional structures grown in laboratories using stem cells or cells taken from human tissues. Depending on the cells used, scientists can grow models that reproduce certain features of organs such as the intestine, liver, brain and lungs. In cancer research, scientists can use cells collected from a patient’s tumour to grow patient-derived tumour organoids. These models retain some of the characteristics of the original cancer and allow researchers to test how different medicines affect it.

Cancer research: Brightfield microscopy image showing three-dimensional patient-derived colorectal cancer tumour organoids grown in a laboratory.
Patient-derived tumour organoids used in cancer research to study tumour behaviour and treatment responses. Image: S.N. Ooft and M. Mertz / The Scientist.

This approach provides more information than experiments using cancer cells grown in a flat layer in laboratory dishes. Since organoids have a three-dimensional structure, they can better represent some of the conditions in which cells grow and interact inside the body.

Researchers are also studying whether patient-derived organoids could help doctors identify treatments that are more likely to work for individual patients. However, this approach still requires further research and is not yet a replacement for established clinical treatment decisions.

Can it Help Cancer Treatment?

The TAVIS project is part of a wider effort to use space-based research to understand human biology and develop medical applications on Earth. After the experiment, researchers will compare the space-grown tumour models with those maintained on Earth. They will examine whether microgravity caused changes in the cancer cells and whether those changes affected their response to chemotherapy.

“If microgravity makes these tumors more responsive, we can look for the genetic changes behind that and try to induce the same changes here on Earth,” said Shay Soker, Ph.D., WFIRM professor and principal investigator of the study. “It’s similar in spirit to immunotherapy, which makes tumors more visible to the immune system. The tumor can run, but it can’t hide.”

If the space-grown models respond differently, scientists will investigate whether the same effects can be produced under normal conditions on Earth. This could help identify new ways to make cancer cells more responsive to existing treatments.

The experiment does not mean that cancer can be treated more effectively in space. Instead, it uses the unusual conditions aboard the ISS to investigate biological changes that are difficult to study in conventional laboratories. By understanding how cancer cells respond to different conditions, researchers hope to gain new insights into tumour behaviour, drug development and the possibility of more personalised cancer treatment.

Continue Reading

Trending