AI Trends in healthcare ppt for bio medial

AI is revolutionizing healthcare by enhancing diagnostics, personalizing treatment, and improving operational efficiency. Advanced algorithms analyze medical images, leading to faster and more accurate diagnoses, particularly for critical conditions like cancer. In the realm of personalized medicine, AI tailors treatment plans based on individual genetic and health data, optimizing outcomes and minimizing side effects. Operationally, AI streamlines administrative tasks such as scheduling and billing, allowing healthcare professionals to devote more time to patient care. Additionally, AI is transforming telehealth by facilitating remote monitoring and virtual consultations, increasing accessibility for patients. As these technologies evolve, ethical considerations and data privacy will be vital to ensure equitable access and maintain trust in AI-driven solutions, ultimately making healthcare more efficient and personalized.

1. 1 Importance in AI Healthcare Dr. D.Sudarvizhi Assistant Professor Sl.G

2. A ge n d a 2 Introduction Drug discovery and molecular modelling Drug Delivery Diagnosis Imaging Surgery P a t i en t M o n i t o r i n g s y s t e m M e d i c a l d e v i c e s D a t a s e c u r i t y a n d pr i v a c y

3. Types of Machine learning and application 3

4. Types of Machine Learning Supervised (inductive) learning Given: training data + desired outputs (labels) T h e l e a rn e r r e c e i v e s a s e t o f l a b e l e d e xa m p l e s as t r a i n i n g data and makes predictions for all unseen points. This is t h e m o s t c o mm o n s c e n a r i o ass o c i a t e d w i t h c l ass i f i c a t i o n , r e g r e ss i o n , a n d r a n k i n g p r o b l e m s . T h e spam detection problem discussed in the previous section is an instance of supervised learning. 4

5. Types of Machine Learning Unsupervised learning – Given: training data (without desired outputs) The learner exclusively receives unlabeled t ra i n i n g da t a, and m a k e s p r e d i ct i o n s f o r a l l u n s ee n p o i n t s . S i n c e i n g e n e ral n o l ab e l e d example is available in that setting, it can be difficult to quantitatively evaluate the p e rf o r m an c e o f a l e ar n e r . Cl us t e r i n g and d ime ns i o na l i t y r e d u ct i o n are e xa m p l e o f unsupervised learning problems. 5

6. Types of Machine Learning Semi-supervised learning – G i v e n : t r a i n i n g d a t a + a f e w d e s i r e d o u t pu t s The learner receives a training sample consisting of both labeled and unlabeled data, and makes predictions for all unseen points. common in settings where unlabeled data is easily accessible but labels are expensive to obtain. Various types of problems arising in applications, including classification, regression, or ranking tasks, can be framed as instances of semi-supervised learning. The hope is that the distribution of unlabeled data accessible to the learner can help him achieve a better performance than in the supervised setting. The ana l y s i s o f t h e c o n d i t i o n s u n d e r w h i c h t h i s c an i n d ee d b e r e a li z e d i s t h e t o p i c o f m u c h m o d e rn t h e o r e t i c al and ap p li e d m a c h i n e l e arn i n g r e s e ar c h. • 6

7. 7 What kind of problems can be tackled using machine learning Predicting the label of a document, also known as document classification, is by no means the only learning task. Machine learning admits a very broad set of practical applications, which include the following: Text or document classification. Assigning a topic to a text or a document, or determining automatically if the content of a web page is inappropriate or too explicit; it also includes spam detection. Natural language processing (NLP) . Most tasks in this field, including part-of speech tagging, named-entity recognition, context-free parsing, or dependency parsing, are cast as learning problems. Speech processing applications. Like speech recognition, synthesis, speaker verification, identification, as well as sub-problems such as language modeling and acoustic modeling. Computer vision applications . Like object recognition, identification, face detection, Optical character recognition (OCR), content-based image retrieval, or pose estimation. Computational biology applications . Includes protein function prediction, molecular modeling,identification of key sites, or the analysis of gene and protein networks.

8. From Data to Model What do we want to predict W ha t da t a c o u l d w e us e t o d o i t Individual Variables R e l a t i o n s h i p b e t wee n v a r i a b l e s M i ss i n g e n t r i e s Outliers S pu r i o u s v a l u e s F o r m u l a t e t h e problem E xp l o re t h e data Clean the data Feature E n g i n e e r i ng C h o o s e & f i t model(s) E v a l ua t e m o d e l performance Construct T r a n sf o rm select Linear T r e e b as e d Neural S u p p o rt ve c t o r K n e a r e s t , B a ye s i an N W Underfit 8 R i g ht o v e rf i t

9. v i c e s | © 2 01 8 9 Algorithm types

10. Machine learning Process 10

11. Modelling lifecycle 11

12. Potential opportunities in Healthcare 12

13. 13 Potential opportunities in Pharmaceuticals M anu f a c t u r i n g P r o c e s s I m p r o v e m e n t : Perform quality control, shorten design time, reduce materials waste, improve production reuse, perform predictive maintenance Reduce material waste, faster production, and more consistently meeting the product’s Critical Quality Attributes (CQAs D r u g D i s c o v e r y & D e s i g n : drug target identification and validation; target-based, phenotypic, as well as multi-target drug discoveries d r u g re pu r p o s i n g ; a n d b io m a r k er i d e n t i f i c a t i o n P r o c e ss i n g B i o m e d i c a l an d cli n ic a l da ta Interpret large volumes of publications and extract text to validate or discard hypothesis Clinical data of patient log in terms of drug and its impact and cross referencing of data and building prediction models Studies show that 80% of clinical trials fail to meet enrolment timelines and one third of study terminations are due to enro lment difficulties R a r e D i s e a s e an d p e r s ona li s e d m e d i c i n e Combining information from body scan, patient biology with AI analytics can detect cancer and even predict health issues IBM WATSON for oncology helps in making patient information, reaction to past drugs to recommend personalised treatment plan I d e n t i f y i n g cli n ic a l t r i a l c and i da t e s Can analyse genetic information to identify right patient population for a trial P r e d ic t i n g t r e a t m e n t r e s u l ts Match drug to patient from information like body's ability to absorb compounds, distribution and metabolism P r e d ic t i v e B i o m a r k e r s Predictive biomarkers are used to identify potential responders to a molecular targeted therapy before the drug is tested in humans D r u g r e pu r po s i n g Repurposing previously known drugs or late-stage drug candidates towards new therapeutic areas is a desired strategy for many biopharmaceutical companies as it presents less risk of unexpected toxicity or side effects in human trials, and, likely, less R&D spend D r u g A dh e r e n c e an d do s a g e Analyse the trial patients drug study protocol so that the correctness of treatment can be ensured

14. Current Healthcare Practice D r i v e r s a n d T r e nd s : M a j o r D em o g rap h i c t rans i t i o n w i t h > 6 5 ye ars ( e xp e ct e d t o r e a c h 16% by 2050) outnumber Children aged 5 years and above. WHO estimate by 2025 , 70% of all illnesses will be chronic, c o m o r b i d c ond i t i on . T o t al G l o ba l h e a l t h c are c o s t fr o m 8 . 4 T n U S D i n 201 5 t o 18 . 3 T n U S D in 2030 Lost productivity in globally in terms of GDP due to illness 47 Tn USD With rise of EMR, personalised genomics, lifestyle, health data and c apa c i t y f o r b e tt e r , fas t e r ana l y s i s o f da t a b y c o rr e l a t i o n ana l y s i s fr o m da t a c o ll e ct e d and h e a l t h c are da t a. 14

15. Current Healthcare Practice D r i v e r s an d T r e nd s : Many Digital stake holders are disrupting by investing … Google building system biology program and applying in pathology Apple, Garmin, Fitbit using sensor data from smart watch to p r e d i c t h e a l t h o f an i nd i v i du a l . Technology and pharma co unusual partnership Google with Novartis, Sanofi with Pfizer, IBM with Merck, Apple w i t h J & J , M S w i t h A s t r a Z e n e c a. P e rs o n a l i s e d M e d i c i n e b y t a i l o r i n g t r e a t me n t 15

16. AI ML/DL It is used to extract structured data from unstructured clinical text data that is often hard for clinicians or professionals to process. NLP is regarded as one of the most developed of the healthcare analytics applications. A la r g e p a rt o f c li n i c a l i n f o rm a t i o n i s i n t h e f o rm o f n a rr a t i ve t e x t f r o m physical examinations, lab reports and discharge summaries. DL a version of ANN Technique like a neural network with many layers of a b s t r a c t io n r a t h er t h a n i np u t a n d o u t p u t . could be feeding the system with millions of x-ray images, each labeled with the desired answer, such as the presence of a nodule/tumor, and once sufficiently trained, the algorithm can easily recognize a potential nodule in an image Precision medicine in three different types of clinical areas complex Alg, d i g i t a l h e al t h a pp , o m i c s b a s ed t e x t . 16

17. Drug Discovery 17 Drug discovery and development is an immensely long, costly, and complex process that can often take more than 10 years from identification of molecular targets until a drug product is approved. Data available assessing drug compound activity and biomedical data in the past few years. Methods such as support vector machines, neural networks, and random forest have all been used to develop models to aid drug discovery since the 1990s. Recently drug compound property and activity prediction, de novo design of drug compounds, drug receptor interactions, and drug reaction prediction. Properties and activity on a drug molecule are important to know in order to assess its behavior in the human body. Machine learning-based techniques have been used to assess the biological activity, absorption, distribution, metabolism, and excretion (ADME) characteristics, and physicochemical properties of drug molecules ( ChEMBL, PubChem). Machine learning has also been implemented to assess the toxicity of molecules, for instance, using DeepTox, a DL-based model for evaluating the toxic effects of compounds based on a dataset containing many drug molecules. Another platform called MoleculeNet is also used to translate two-dimensional molecular structures into novel features/descriptors, which can then be used in predicting toxicity of the given molecule.

18. Drug Discovery & Development 18

19. Generation of new chemical structures through neural networks, includes protein engineering involving the molecular design of proteins with specific binding or functions. Autoencoders are a type of neural network for unsupervised learning and are also the tools used to, for instance, generate images of fictional human faces. The autoencoders are trained on many drug molecule structures and the latent variables are then used as the generative model(druGAN). Assessment of drug target interactions is an important part of the drug design process. The binding pose and the binding affinity between the drug molecule and the target have an important impact on the chances of success based on the in silico prediction. Some of the more common approaches involve drug candidate identification via molecular docking, for prediction and preselection of interesting drug target interactions. Molecular docking is a molecular modeling approach used to study the binding and complex formation between two molecules. It can be used to find interactions between a drug compound and a target, for example a receptor, and predicts the conformation of the drug compound in the binding site of the target ( FlexX, AutoDock, DOCK, Glide) DL models take a long time to train because of the large datasets and the often large number of parameters needed 19 Drug Discovery & Development

20. Drug Delivery T h e d i s c i p li n e o f m o du l a t i n g m o l e c u l ar f e a t u r e s t o o b t a i n a u g m e n t e d ph y s i c o c h e m i c al p r o f i l e s f o r ph a r m a c o l o g i c al application is generally referred to as drug delivery which represents a multidisciplinary development of the traditional drug formulation sector. Nanotechnology has been a hot topic in drug delivery, as engineering of nanosized matter was found to be particularly well-suited for interacting with the human body and controlling drug distribution. Adaptive algorithms, such as neural networks, have become increasingly relevant in proteomics and bioinformatics, galvanized by the vast genomics and proteomics projects. 20

21. Diagnosis 21 The capacity of artificial intelligence (AI) to predict disease risk, diagnose illness, and guide therapeutic decision making is now a realistic prospect in the foreseeable future, Variety of fields in medicine including cardiology, renal medicine, critical care, and mental health. In the case of solid organ tumors, criteria that determine patient prognosis and management relate to a variety of factors, such as histological tumor grade, tumor stage, and the presence/absence of prognostically relevant features such as vascular and/or lymphatic invasion. The ideal computational pipeline under these circumstances would be capable of seamlessly integrating data from multiple heterogeneous sources (molecular/ clinicopathological/radiological/exposomal) to arrive at a given answer(risk prediction/cancer diagnosis/cancer prognostication/determination of therapeutic efficacy) one can consider the key data challenges in cancer medicine as (1) logistical, (2) precision centered, and (3) next generation AI has huge potential in various phase of Diagnosis like in case of cancer, susceptibility, diagnosis and staging, predict treatment response, predict recurrence and survival, personalised cancer pharmacotherapy ( drug selection, toxicity prediction, drug pairing, drug repurposing)

22. I m a g i n g 22 Computer vision involves the interpretation of images and videos by machines at or above human-level capabilities including object and scene recognition. Areas where computer vision is making an important impact include image-based diagnosis and image-guided surgery. DL is used to engineer computer vision algorithms for classifying images of lesions in skin and other tissues. Video data is estimated to contain 25 times the amount of data from high resolution diagnostic images such as CT and could thus provide a higher data value based on resolution over time. Video analysis of a laparoscopic procedure in real time has resulted in 92.8% accuracy in identification of all the steps of the procedure and surprisingly, the detection of missing or unexpected steps. Convolving an image with various weights and creating a stack of filtered images is referred to as a convolutional layer, where an image essentially becomes a stack of filtered images. Pooling is then applied to all these filtered images, where the original stack of images becomes a smaller representation of themselves and all negative values are removed by a rectified linear unit (ReLU)

23. I m a g i n g DICOM (Digital Imaging and Communications in Medicine) is the standard protocol for managing and communicating the medical image information and related data. A DICOM file consists of site of origin, patient identification, the image itself, and attributes of the image such as pixel size. DICOM files insure that patient data and picture data cannot be separated. Consequently, the DICOM image is always linked to the patient. One of the biggest hurdles in using AI in radiology is the need for a vast amount of high-quality labeled datasets that have a satisfactory training model, a balanced dataset that is representative of all data. 23

24. I m a g i n g 24 RadBot-CXR has been developed and validated to an expert level automatic interpretation system for the detection of four categories that relates to seven distinct radiographic findings on CXR: alveolar consolidation, lung mass, atelectasis, pleural effusion, hilar prominence, diffuse pulmonary edema, and cardiomegaly. The four main categories are general opacity, cardiomegaly, hilar prominence, and pulmonary edema, where the software output reports will verify the existence or the nonexistence of each of these categories Detecting osteoporis ZMV team were able to obtain segments of patches of the vertebral column. These patches were binary classified, put together, and trained using a CNN and then run on recurrent neural networks (RNNs). Butterfly iQ is an innovative pocket-sized ultrasound device, developed by the Butterfly Network. It utilizes capacitive micromachined ultrasound transducer (CMUT)/complementary metal oxide semiconductor- based parts in the ultrasonic probe. AI in medical imaging of arteries which can support cardiovascular disease and blood vessel diagnosis in terms of blood flow, by a single, short scan, compared to a conventional scan, 4D flow acquires volumetric anatomical, functional, and flow information during the entire c a r d ia c c y c l e . The model has also been extrapolated to analyze the lungs, liver, and CXR, while developing neuro and prostate platforms, creating a healthcare AI suite

25. S u r g e r y 25 Ongoing transformation within surgical technology and focus has especially been placed in reducing the invasiveness of surgical procedure by minimizing incisions, reducing open surgeries, and using flexible tools and cameras to assist the surgery. Minimal invasive surgery requires different motor skills compared with conventional surgery due to the lower tactile feedback when relying more on tools and less on direct touching. Sensors that provide the surgeon with finer tactile stimuli are under development and make use of tactile data processing to translate the sensor input into data or stimuli that can be perceived by the surgeon. Such tactile data processing typically makes use of AI, more specifically artificial neural networks to enhance the function of this signal translation and the interpretation of the tactile information. Artificial tactile sensing offers several advantages compared with physical touching including a larger reference library to compare sensation and standardization among surgeons with respect to quantitative features, continuous improvement, and level of training. artificial tactile sensing has been used includes screening of breast cancer, as a replacement for clinical breast examination to complement medical imaging techniques such as x-ray mammography and MRI. Artificial tactile sensing has also been used for other applications including assessment of liver, brain, and submucosal tumors

26. 26 S u r g e r y Before proceeding with an operation, the surgeon must determine if the patient can tolerate the procedure. like adequate cardiac function Laparoscopic surgery continues to present numerous technical challenges, however, with limited degrees of freedom for instrument articulation, diminished tactile feedback, and limited 2D vision, which makes technically challenging procedures more difficult except for t h e m as t e r l a p a r o s c o p i c s u r g e o n s. C o m pu t e r - ass i s t e d l apar o s c o p i c sur g e ry o f f e rs au g me n t e d capabilities to address some of these issues. The da Vinci robotic system (Intuitive Surgical Inc) offers a “master- slave” configuration with the surgeon at a bedside console using inputs to control the robotic arms inside the patient. Its display gives a t h r e e - d i me n s i o n al i m a g e w i t h i n s t r u me n t s t h at h a v e i n c r e as e d degrees of freedom to address the prior listed shortcomings of l a p a r o s c o p i c s u r g e r y .

27. 27 Patient Monitoring There are many areas in healthcare in which NLP can provide substantial benefits. Some of the more immediate applications include Efficient billing: extracting information from physician notes and assigning medical codes for the billing process. Authorization approval: Using information from physician notes to prevent delays and administrative errors. Clinical decision support: Facilitate decision-making for members of healthcare team upon need (for instance, predicting patient prognosis and outcomes). Medical policy assessment: compiling clinical guidance and formulation appropriate guidelines for care. Disease classification based on medical notes and standardized codes using International Statistical Classification of Diseases and Related Health Problems DeepCare is an example of an AI-based platform for end-to-end processing of EMR data. It uses a deep dynamic memory neural network to read and store experiences and in memory cells. Using the stored data, the framework of DeepCare can model disease progression, support intervention recommendation, and provide disease prognosis based on EMR databases.

28. Patient Monitoring The various devices now used in healthcare are able to capture huge volumes of data for analysis and the emergence of the Internet of Things (IoT) has enabled devices to be linked to one another, around the home and environment. Diseases that were traditionally difficult for clinicians to manage because they relied heavily on patient compliance, can now be managed using AI technology. Activity trackers in every device are now commonplace and they generate huge amounts of data which are then used to develop new tools. AI application in adults whose breathing at night compromises their airway (obstructive sleep apnea, OSA) are shown to have higher degradation in areas of the brain, which are known to be linked to dementia with sensors implanted in pillows help better management. AI can be combined with existing technology such as electrodes to measure or stimulate brain activity (electromyography, EMG) or cameras to assess the type of hand grip required to clasp an object and allows better control of robotics such as prosthetic limbs. Deep brain stimulation has grown in use as the benefits in recalcitrant patients are seen and the processes to offer the procedure become more widely available—it is used for movement disorders and is also being investigated for psychiatric d i s o r d e rs.

29. Medical Devices – Bionic hand T h e pa i r i n g o f a ha r d w are ( c a me ra) w i t h a s o f t w are algorithm has also allowed a bionic hand to adapt the kind of grip it uses to clasp different objects that the camera has been trained to recognize. The bionic limbs are also being developed to provide feedback to users, allowing them to have a p e r c e p t i o n o f s e nsa t i o n — i n t h e h a nd , s e n s o rs o f f e r kinesthetic feedback to amputees so that they can do more than just provide movement but sensation t oo , t h ro u g h w h i c h t h e y c an i m p ro v e m o t o r c o n t r o l 29

30. Medical Devices Wearable devices such as wrist sensors are common nowadays and have many of the same sensors as smartphones. They can be used to detect motion such as those associated with smoking and seizure activity Wearable sensors are also used by clinicians, for instance, wearable patches can measure muscle activity and posture , radiofrequency sensors placed over clothes can detect pulmonary function , and others can assess the recovery process of several neurological disorders . There is an emerging technology that consists of a miniature sensor implanted in a pill, which emits signals to a wearable p a tc h w h e n i t e n t e rs t h e s t o m a c h . Major application of AI in cognitive/mental health monitoring consists in Chatbot coaches. Mobile health apps such as WoeBot, Youper, Wysa, Replika, and Unmind provide conversational AI Chatbots that offer day-to-day emotional support and me n t al h e a l t h t r a c k i n g t h r o u g h s m a r t ph o n e s 30

31. Data Security and Privacy The increasing diversity among healthcare providers adds to the complexity of protecting data security and privacy in health, no less because of the disruptive impact new technologies, including AI, have on the industrial organization 31

32. Thank You 32