Kwang Gi Kim

Plain English
This study developed an AI-based system to help diagnose a condition in infants called developmental dysplasia of the hip (DDH) using ultrasound. The new system was highly accurate, with the best model showing an Area Under the Curve score of 0.864, which means it was very effective in identifying DDH compared to expert doctors. This innovation is important because it can make hip assessments in babies quicker and more accessible, potentially improving early diagnosis and treatment. Who this helps: This helps infants and their healthcare providers by providing faster and more reliable diagnoses.

Plain English
This study looked at how silicon-based materials used in lithium-ion batteries change over time and what causes them to degrade. Researchers created a digital twin model to better understand the mechanics behind these changes and found that charging the battery faster (from 0.5C to 4C) helped reduce stress within the battery, leading to less damage and improved stability. This matters because it can lead to the development of longer-lasting, more efficient batteries. Who this helps: This benefits battery manufacturers and consumers looking for more reliable energy storage solutions.

Plain English
This study looked at a new computer model called U-Net3+ that helps detect bleeding in the eyes caused by diabetic retinopathy, a complication of diabetes that can lead to blindness. The model performed very well, achieving 99.93% accuracy in identifying hemorrhages, 87.03% sensitivity in correctly detecting actual cases, and 99.97% specificity in reducing false alarms, especially in areas with more severe bleeding. This is important because early detection can prevent serious vision loss for people with diabetes. Who this helps: This benefits patients with diabetes and their doctors by improving early screening for eye problems.

Plain English
This study looked at a new sensor designed to monitor the flow of fluids during a medical procedure called paracentesis, which is often performed on cancer patients. The sensor, which is attached to a drainage tube, can detect when liquid is flowing by measuring a tiny voltage of 11.07 microvolts created by the movement of fluid. This non-invasive technology not only improves safety and reduces infection risk but also cuts down on costs since it can be reused. Who this helps: This benefits cancer patients undergoing paracentesis and the doctors performing the procedure.

Plain English
This study looked at patients who had a specific type of surgery called pancreaticoduodenectomy and how likely they are to develop new diabetes within three years. Researchers found that about 37% (47 out of 126 patients) went on to develop diabetes after surgery. They created a more accurate prediction model using a combination of CT scan images and clinical data, which performed significantly better than older methods. Who this helps: This benefits patients undergoing pancreatic surgery by providing better risk assessments for diabetes.

Plain English
This study explored whether a special type of MRI, called noncontrast abbreviated MRI (NC-AMRI), combined with machine learning, could better tell apart liver cancer (hepatocellular carcinoma, HCC) from benign growths (dysplastic nodules, DNs). Out of 189 patients, the machine learning models achieved accuracy scores (AUC) ranging from 0.758 to 0.798, meaning they successfully identified HCC in many cases while also showing strong results for distinguishing more concerning DNs. This research is important because it improves the ability to identify serious liver conditions without needing contrast dye, making the process safer and more accessible. Who this helps: This benefits patients undergoing liver evaluations and doctors diagnosing liver conditions.

Plain English
This study looked at a new system that helps surgeons find lymph nodes during cancer surgeries using a special dye and light. The system uses a photosensor that reacts to the dye by lighting up and sounding an alarm when lymph nodes are detected, making it easier and quicker for surgeons to locate them. This method could significantly cut down on surgery time, which is important for patient recovery. Who this helps: Patients undergoing cancer surgery.

Plain English
This study looked at how well two deep learning models, U-net and U-net3+, could detect damage from diabetic retinopathy in different parts of the retina. The U-net3+ model was better at identifying these lesions, achieving a high score of about 88% accuracy in important areas of the retina. This research is significant because improved detection can lead to better diagnoses and treatments for people with diabetic retinopathy. Who this helps: Patients with diabetic retinopathy.

Plain English
This study focused on improving breast cancer detection using a new technology called PMVnet, which analyzes paired mammogram images instead of just one. In tests with 1,636 mammograms, PMVnet showed it could better identify breast lesions, achieving a high recall rate of 95% and reducing false positives significantly compared to traditional methods. This matters because it can help doctors find breast cancer more accurately, leading to earlier detection and better treatment outcomes for patients. Who this helps: Patients undergoing mammograms for breast cancer screening.

Plain English
This study looked at whether advanced computer programs, known as deep learning models, can predict if dental implants can be placed right after tooth extraction using panoramic X-rays. Researchers evaluated data from 201 patients and found that all the models tested performed well, scoring over 90% in key measures like accuracy and precision. This is important because it could help dentists make better decisions about implant placement, potentially improving patient outcomes and streamlining the extraction process. Who this helps: This benefits dentists and their patients who need dental implants.

Plain English
This study developed a new technology using artificial intelligence to improve the images seen during cataract surgery. The system automatically generates high-quality images by training on existing pictures, leading to more accurate visualization for surgeons. The technology was tested and showed excellent performance with measures indicating a strong ability to produce optimized images (with scores like PSNR of 29.887). Who this helps: This helps cataract surgery patients by providing surgeons with better visual tools to enhance the surgery's success.

Plain English
This study looked at whether chest CT scans can be used to diagnose sarcopenia, which is a loss of muscle mass and function. Researchers analyzed 4,932 chest CT scans from 1,644 patients and found that an artificial intelligence model called UNet3+ accurately measured the pectoralis muscle area, achieving a performance score of 0.95. The results showed a strong correlation between the pectoralis muscle area and the critical muscle area measured at L3, suggesting that chest CT can effectively help diagnose sarcopenia. Who this helps: This benefits patients at risk of sarcopenia and their doctors by providing better diagnostic options.

Plain English
This study focuses on a new method for improving the accuracy of medical data, specifically waveform data like electrocardiograms (ECGs), by using graph structures. The researchers found that their approach increased the accuracy of predictions by 1.44% and made models stronger against attacks, with an overall performance improvement of 2.47% when combined with other techniques. This is important because it helps ensure that medical diagnoses based on these waveforms are more reliable. Who this helps: This benefits doctors and medical professionals who rely on accurate ECG readings for patient care.

Plain English
This study looked at how artificial intelligence (AI) can help doctors accurately measure the size of livers in living donors before and after surgery. The researchers analyzed the CT scans of 55 donors and found that the AI models were highly accurate, achieving a score of 95.73% for pre-surgery images and showing a consistent ability to measure liver size changes, with an average liver resection rate of 40.52% and a regeneration rate of 13.50% by about two months after surgery. This information is important because it helps ensure that donors recover well and that transplant procedures are safer and more effective. Who this helps: This benefits patients undergoing liver donation and the medical teams managing their care.

Plain English
This research focused on a process called calendering, which shapes battery electrodes to improve their performance. The study found that by using advanced imaging and simulations, they could accurately track changes in the electrode's structure during manufacturing, leading to better battery efficiency. This is important because it helps create more effective batteries that can store more energy and last longer. Who this helps: Patients and consumers, especially those using devices powered by lithium-ion batteries.

Plain English
This study looked at how artificial intelligence (AI) can help detect esophageal cancer earlier by analyzing endoscopy images. The researchers tested two AI models, finding that one model (RetinaNet) was very accurate, detecting 98.4% of actual cases while missing only 8.7% of the time, and the other model (YOLOv5) had similar success, with a detection rate of 93.7% and a slight miss rate of 10.1%. This matters because earlier detection could lead to better treatment outcomes and reduce the likelihood of misdiagnosis for patients with esophageal cancer. Who this helps: This benefits patients at risk for esophageal cancer and the doctors treating them.

Plain English
This study looked at what factors might make pneumonia patients more likely to return to the hospital within 30 days of being discharged. Researchers analyzed data from 22,600 pneumonia patients and found that among various risk factors, things like age, red blood cell distribution width, and cancer history were the most important in predicting readmission. Interestingly, the amount of greenery in a patient's neighborhood also played a role, ranking as the 15th most important factor, showing that living near more plants and trees might help lower readmission rates. Who this helps: This research helps doctors and healthcare providers identify patients at higher risk for readmission, particularly those with pneumonia.

Plain English
This study looked at how to predict whether patients with worsening chronic respiratory diseases might die while in the hospital. Researchers examined data from over 6,200 patients and identified several important factors that affect mortality risk. They found that things like high blood urea nitrogen levels and older age were linked to higher risk, while good albumin levels were associated with lower risk. The model they created was very effective, with a prediction accuracy rate of around 94% for the worst outcomes, highlighting the impact of long-term air pollution exposure on these patients. Who this helps: This benefits patients with chronic respiratory diseases and the doctors who treat them.

Plain English
This study focused on using artificial intelligence to analyze knee X-ray images to better diagnose a condition called subchondral sclerosis, which is related to osteoarthritis. The researchers tested various AI models on 4,019 knee images and found that the best model accurately identified the condition with 84.7% accuracy and 92.5% specificity. This advancement matters because it can lead to more accurate and consistent diagnoses, helping doctors make better treatment decisions. Who this helps: This helps patients with osteoarthritis, doctors diagnosing the condition, and healthcare systems aiming for improved care.

Plain English
This research studied how well a machine learning technique called ensemble learning can improve the early diagnosis of Alzheimer's disease (AD) by analyzing brain activity and results from a clock drawing test. The study found that using a combination of data from brain scans (EEGs) and clock drawings led to a 20% increase in the accuracy of identifying Alzheimer's compared to using each type of data alone. This is important because better detection of Alzheimer's can lead to earlier treatment and better outcomes for patients. Who this helps: This helps patients and their families by enabling quicker and more accurate diagnoses of Alzheimer’s disease.

Plain English
This study looked at how artificial intelligence can help emergency departments quickly assess the severity of patients based on their conversations. Researchers analyzed 1,028 transcripts of discussions between medical staff and patients and found that their AI model could categorize patients effectively, achieving a score that indicates good accuracy (0.764). This is important because it can lead to faster treatment for patients, reduce wait times, and help manage crowded emergency rooms better. Who this helps: This helps patients who need timely care in emergency situations.

Plain English
This study looked at how a deep learning algorithm can improve the visibility of images used in a specific type of eye surgery called vitreoretinal surgery. Researchers found that the algorithm significantly enhanced the quality of surgical images, achieving high scores for clarity and similarity to real life, with improvements in sharpness, brightness, and color contrast. This matters because better visual quality during surgery can help doctors perform procedures more effectively and safely. Who this helps: Patients undergoing vitreoretinal surgery.

Plain English
This study looked at using artificial intelligence (AI) to help diagnose knee osteoarthritis (KOA), a painful condition that affects 37% of people over 60. Researchers trained AI models on 15,000 images to automatically rate the severity of KOA using a standard grading system. The best model, DenseNet201, correctly identified the severity of KOA 73% of the time and performed especially well with higher grades, achieving 82.7% accuracy for the most severe cases. Who this helps: This helps doctors diagnose knee osteoarthritis more accurately and quickly.

Plain English
This study focused on using artificial intelligence to identify the specific causes of speech disorders in people who have suffered from strokes or hearing impairments. The researchers developed a deep learning model that analyzed voice recordings and found that their models could accurately classify these disorders, with scores showing promising accuracy: 83.9% for one model, 91.3% for another, and 90.6% for a third. This is important because it could lead to quicker and more precise diagnoses for patients with speech disorders, helping to guide their treatment effectively. Who this helps: Patients with speech disorders and their healthcare providers.

Plain English
This study looked at how well advanced AI systems, specifically large language models like ChatGPT4 and Claude3-Opus, could create and answer questions about oral and maxillofacial surgery. The models were able to answer over 90% of the questions correctly, with a notable 97% accuracy for questions that included images, compared to 88.9% for text-only questions. This is important because it shows that while these AI tools can be useful for generating medical information, they still have limitations, especially in fully understanding context and providing accurate answers to all generated questions. Who this helps: This helps doctors and medical professionals looking for reliable AI support in oral and maxillofacial surgery.

Plain English
This study looked at how well different computer programs could tell if vertebral compression fractures (VCFs) on CT scans are benign (non-cancerous) or malignant (cancerous). Researchers examined CT data from 447 fractures and found that a deep learning model outperformed traditional machine learning methods, achieving a slight edge with an accuracy rate of 77.66%, compared to 75.91% for the best machine learning approach. This is important because accurately distinguishing between these types of fractures can significantly impact patient treatment plans and outcomes. Who this helps: This helps doctors in making better diagnoses for patients with vertebral compression fractures.

Plain English
This study looked at using advanced computer technology to analyze walking patterns in patients with incomplete spinal cord injuries and cauda equina syndrome. Researchers collected walking data from 214 patients and found that a specific machine learning model accurately classified these conditions 74.42% of the time. Understanding these walking issues better can lead to earlier diagnoses and tailored treatments, improving the quality of life for these patients. Who this helps: Patients with incomplete spinal cord injuries and cauda equina syndrome.

Plain English
This study looked at how well different types of brain images (DWI and ADC) can identify areas of damage in stroke patients using artificial intelligence. They tested these images from 360 patients and found that the U-Net model was more effective at segmenting lesions, showing an accuracy rate of 92.13% for DWI images compared to 83.68% for ADC images. Understanding which imaging type performs better is important because it can help doctors diagnose strokes more accurately and quickly. Who this helps: Patients who have suffered a stroke and need timely and accurate diagnosis.

Plain English
This study developed a system that uses advanced computer technology to automatically sort and identify various types of pills. Researchers tested 4080 images of 102 different pills and achieved a high accuracy rate of 88.8% in identifying them. This matters because it could improve the efficiency of pill sorting in pharmacies and hospitals, leading to better medication management for patients. Who this helps: This helps pharmacists and healthcare providers.

Plain English
Researchers studied how well different computer programs can automatically grade cervical cancer by analyzing images of areas that appear white after applying a special solution. They looked at 464 cervical images and found that the programs performed better when they included a small area around these white lesions. For instance, one program (SVM) achieved an accuracy of 87% when it considered this extra margin, compared to 83% without it, showing that this approach can improve cancer diagnosis. Who this helps: This helps doctors better diagnose cervical cancer in women.

Plain English
This study compared three types of advanced computer networks—convolutional neural networks (CNNs), vision transformers (ViTs), and hybrid networks—to see which could best identify and outline the paranasal sinuses in people with sinusitis using CT scans. The hybrid networks, especially one called Swin UNETR, performed the best, achieving a Jaccard Index score of 0.719, which indicates strong accuracy, while also being efficient in terms of processing time and complexity. This matters because better segmentation can improve surgical safety by helping doctors navigate key anatomical features during surgery for sinusitis. Who this helps: This benefits patients undergoing treatment for sinusitis and the doctors performing their surgeries.

Plain English
This study focused on using artificial intelligence to measure liver volumes in patients before and after liver surgery. The AI model achieved a high accuracy rate in measuring liver size, with scores of around 94% in different time periods after surgery. This technology is important because it can help doctors better plan surgeries and monitor how well patients recover after their operations. Who this helps: This benefits patients undergoing liver surgery and their doctors.

Plain English
This study focused on creating a computer program using deep learning to automatically measure specific angles in the knee from CT scans, which is important for diagnosing and treating conditions like osteoarthritis. The program was tested on CT scans from 270 older adults and showed high accuracy—over 92% in identifying relevant features and making measurements that closely matched those done by human radiologists. This automated approach also saved time, reducing measurement duration by about 28% compared to manual methods, which is crucial in busy medical environments. Who this helps: This helps doctors and radiologists by making their work faster and more efficient, ultimately benefiting patients with knee issues.

Plain English
This study looked at how artificial intelligence (AI) can help dentists predict the number of dental implants needed for patients with missing teeth. Researchers developed a model that analyzed data from 628 patients and was able to accurately predict implant counts, achieving a very low error rate with measurements indicating strong reliability (with an error score of 0.0871). This is important because it can lead to better treatment planning, making procedures more consistent and improving overall patient care. Who this helps: This helps dentists, especially those who may be less experienced with implant planning.

Plain English
This study looked at the presence and variety of Bartonella bacteria in wild rodents and their fleas in South Korea. Researchers found that nearly half of the rodent samples tested positive for Bartonella, with specific species like Bartonella grahamii present, which can make people sick. This is important because it highlights the role of these wild rodents and their fleas in spreading diseases that can affect humans, emphasizing the need for careful monitoring to prevent potential infections. Who this helps: This helps patients, doctors, and public health officials by raising awareness about zoonotic diseases.

Plain English
This study looked at how well different YOLO-based deep learning models could automatically identify two types of spinal issues—osteophytes and ossification of the posterior longitudinal ligament (OPLL)—using cervical CT scans. Out of nearly 2,700 images, the models found that 79.5% had both conditions, with the YOLOv5 model performing the best, achieving a precision of 67.42% and a recall of 68.36%. These results are important because accurate detection helps doctors make better decisions about surgeries like artificial disc replacement. Who this helps: This helps patients who may need spinal surgery.

Plain English
This study focused on using artificial intelligence (AI) to accurately classify different types of bladder tumors from digital images of tissue samples. Researchers trained AI models on 12,500 images and found that the best model, called EfficientNet-B6, had a high accuracy of 91.3%, meaning it correctly identified tumors most of the time. This is important because it could lead to faster and more accurate diagnoses of bladder cancer, helping improve patient outcomes. Who this helps: Patients with bladder cancer and their doctors.

Plain English
This study explored how well different artificial intelligence models could identify and outline important female reproductive organs during minimally invasive gynecological surgery, using images from 21 patients. The U-Net model was the best performer for identifying the uterus with a high accuracy score, while all models struggled significantly with smaller organs like the ovaries and fallopian tubes. These findings reveal that while AI shows potential in surgical imaging, there are still major challenges to overcome, especially when it comes to accurately segmenting smaller, harder-to-see structures. Who this helps: This helps surgeons and medical professionals improve surgical planning and procedures for gynecological patients.

Plain English
This study focused on improving the decision-making process for when to remove ureteral stents, which are small tubes placed to help urine flow from the kidneys. Researchers developed a new imaging analysis method that objectively evaluates changes in the stents after surgery. They found that their analysis showed significant changes in the ureters over time, which can help determine the right time for stent removal, improving patient care. Who this helps: This benefits patients needing stent removal and their doctors by providing clearer guidelines for timely action.

Plain English
This study looked at how a deep learning model using data from a 24-hour Holter-ECG (a type of heart monitor) can predict serious heart problems in patients who have had heart failure or a heart attack. Researchers followed over 1,100 patients and found that the model successfully identified risks, showing that those in the high-risk group had a seven-fold higher chance of dying from heart issues compared to the low-risk group. This is significant because using this model can help doctors better identify patients who need closer monitoring and care. Who this helps: This helps patients with heart failure or a history of heart attacks, as well as their doctors.

Plain English
This study looked at how a specialized rehabilitation robot can help patients who have shoulder pain due to a stroke and a condition known as frozen shoulder. The researchers found that the robot's movements were very accurate, with positional errors between 0.5% and 2.8%, which means it closely mimicked the natural motion of a healthy shoulder. This is important because improving shoulder movement can help stroke patients regain their ability to perform everyday tasks more easily. Who this helps: This benefits stroke patients experiencing shoulder pain and physical therapists using rehabilitation robots.

Plain English
This study evaluated a new method for delivering medication to treat primary malignant bone tumors using steam. The researchers found that their steam delivery system heated the target area in the bone to about 48.8°C, allowing for better distribution of the treatment compared to standard approaches. This is important because it may improve outcomes for patients suffering from painful bone tumors, making treatments more effective with less damage to surrounding healthy tissue. Who this helps: Patients with malignant bone tumors.

Plain English
This study focused on using advanced computer programs to automatically identify and measure specific thigh muscles from CT scans. Researchers tested three different deep learning models on a total of 176 CT scans and found that all models accurately segmented the muscles, with the best model achieving over 96% accuracy. This new method can help doctors assess muscle health more efficiently, which is important for diagnosing conditions like sarcopenia that impact strength and mobility. Who this helps: This helps patients with muscle-related health issues and their doctors.

Plain English
This study developed a computer program using deep learning technology to measure the size of the remaining pancreas after surgeries called pancreaticoduodenectomy and distal pancreatectomy. It analyzed 1,579 CT scans from 525 patients and found that the best model could accurately measure the remnant pancreas, achieving a success rate of about 77% to 81%. This is important because it helps doctors assess how well the remaining pancreas is functioning after surgery, which can guide patient treatment and management. Who this helps: Patients recovering from pancreatic surgery.

Plain English
This study compared two methods for analyzing images from sleep endoscopy, which helps identify blockages in the airway for people with sleep apnea. It found that centralized learning (CL) outperformed federated learning (FL) in accurately identifying key areas in these images: CL achieved an accuracy score of about 86% for one area (the velum) and 87% for another (the OTE), while FL scored about 82% and 85%, respectively. Improving these techniques is important for creating better tools to help diagnose and treat obstructive sleep apnea. Who this helps: This benefits patients with sleep apnea and their doctors.

Plain English
This study looked at how to diagnose depression using data from people’s sleep patterns collected by wearable devices, rather than relying solely on interviews or questionnaires. The researchers developed a deep learning model that analyzes sleep data and found it could effectively identify depression, achieving a high accuracy score of 0.91. This is important because it offers a more objective and affordable way to diagnose depression, which could lead to better treatment and support for those affected. Who this helps: This helps patients by providing a more accurate and accessible way to diagnose depression.

Plain English
This study looked at how well artificial intelligence (AI) can analyze videos from endoscopic ultrasounds (EUS) to diagnose gallbladder polyps, which are growths in the gallbladder. The best AI models were able to identify polyp regions with over 99% accuracy and classify them as benign or cancerous with an accuracy of about 88% on new data. This matters because using AI can make diagnosing these polyps more accurate and less dependent on the skill of the operator, potentially leading to better patient outcomes. Who this helps: Patients with gallbladder polyps and doctors who diagnose them.

Plain English
The study focused on finding a quick and affordable way to diagnose strokes using speech data from patients who have trouble speaking, a condition called dysarthria. The researchers created a computer model that uses speech patterns to identify stroke symptoms, achieving impressive accuracy rates—96.77% for identifying those who have strokes and 96.08% for correctly determining who does not. This method is important because it allows for faster diagnosis without invasive procedures, potentially leading to better treatment outcomes. Who this helps: This helps patients who may be experiencing strokes and need immediate diagnosis.